Practical speed meter designs for QND gravitational-wave interferometers

Patricia Purdue and Yanbei Chen

Theoretical Astrophysics, California Institute of Technology, Pasadena, California 91125
(October 25, 2018)

In the quest to develop viable designs for third-generation optical interferometric gravitational-
wave detectors (e.g., LIGO-III and EURO), one strategy is to monitor the relative momentum
or speed of the test-mass mirrors, rather than monitoring their relative position. A previous paper
analyzed a straightforward but impractical design for a speed-meter interferometer that accomplishes
this. This paper describes some practical variants of speed-meter interferometers. Like the original
interferometric speed meter, these designs in principle can beat the gravitational-wave standard
quantum limit (SQL) by an arbitrarily large amount, over an arbitrarily wide range of frequencies.
These variants essentially consist of a Michelson interferometer plus an extra “sloshing” cavity
arXiv:gr-qc/0208049v1 17 Aug 2002

that sends the signal back into the interferometer with opposite phase shift, thereby cancelling the
position information and leaving a net phase shift proportional to the relative velocity. In practice,
the sensitivity of these variants will be limited by the maximum light power Wcirc circulating in the
arm cavities that the mirrors can support and by the leakage of vacuum into the optical train at
dissipation points. In the absence of dissipation and with squeezed vacuum (power squeeze factor
e−2R ≃ 0.1) insertedqinto the output port so as to keep the circulating power down, the SQL can be
SQL −2R
beat by h/hSQL ∼ Wcirc e /Wcirc at all frequencies below some chosen fopt ≃ 100 Hz. Here
SQL
Wcirc ≃ 800kW(fopt /100Hz)3 is the power required to reach the SQL in the absence of squeezing.
[However, as the power increases in this expression, the speed meter becomes more narrow band;
additional power and re-optimization of some parameters are required to maintain the wide band.
See Sec. III B.] Estimates are given of the amount by which vacuum leakage at dissipation points
will debilitate this sensitivity (see Fig. 12); these losses are 10% or less over most of the frequency
range of interest (f > ∼ 10 Hz). The sensitivity can be improved, particularly at high freqencies, by
using frequency-dependent homodyne detection, which unfortunately requires two 4-kilometer-long
filter cavities (see Fig. 4).

PACS numbers: 04.80.Nn, 95.55.Ym, 42.50.Dv, 03.67.-a

1
 I. INTRODUCTION 50
SVPM

[Sh (f)/SSQL (100 Hz)]1/2

CP
M
This paper is part of the effort to explore theoreti-
10
cally various ideas for a third-generation interferometric
gravitational-wave detector. The goal of such detectors 5
is to beat, by a factor of 5 or more, the standard quantum SIS
M
limit (SQL)—a limit that constrains interferometers [1]
1 SV
such as LIGO-I which have conventional optical topology SM
0.5 SM
[2,3], but does not constrain more sophisticated “quan- SV
tum nondemolition” (QND) interferometers [4,5]. SQ
The concepts currently being explored for third- L
generation detectors fall into two categories: external 5 10 50 100 500 1000
readout and intracavity readout. In interferometer de- f (Hz)
signs with external readout topologies, light exiting the FIG. 1. Comparison of noise curves (with losses) of sev-
interferometer is monitored for phase shifts, which indi- eral interferometer configurations. Each of these curves has
cate the motion of the test masses. Examples include been optimized in a way that is meant to illustrate their rel-
conventional interferometers and their variants (such as ative advantages and disadvantages. The conventional posi-
LIGO-I [2,3], LIGO-II [6], and those discussed in Ref. [7]), tion meter (CPM) [7] has Wcirc = 820 kW and bandwidth
as well as the speed-meter interferometers discussed here γ = cT /4L = 2π × 100 Hz. The squeezed-input speed me-
and in a previous paper [8]. In intracavity readout ter (SISM)—optmized to agree with the conventional po-
topologies, the gravitational-wave force is fed via light sition meter at high frequencies—has power squeeze factor
pressure onto a tiny internal mass, whose displacement e−2R = 0.1, optimal frequency ωopt = 2π × 105√Hz, extrac-
tion rate δ = 2ωopt , and sloshing frequency Ω = 3ωopt . The
is monitored with a local position transducer. Exam-
squeezed-variational position meter (SVPM) [7] has the same
ples include the optical bar, symphotonic state, and op-
parameters as the conventional position meter, with power
tical lever schemes discussed by Braginsky, Khalili, and
squeeze factor e−2R = 0.1. There are two squeezed-variational
Gorodetsky [9–11]. These intracavity readout interfer- speed-meter curves (SVSM). One (black dashes) uses the
ometers may be able to function at much lower light pow- same parameters as the squeezed-input speed meter. The
ers than external readout interferometers of comparable other (solid curve) has been optimized to compare more di-
sensitivity because the QND readout is performed via the rectly with the squeezed-variational position meter; it has
local position transducer (perhaps microwave-technology Ω = 2π × 95 Hz and δ = 2π × 100 Hz (note that our δ is
based), instead of via the interferometer’s light; however, equivalent to the bandwidth γ used to describe the interfer-
the designs are not yet fully developed. ometers in Ref. [7]).
At present, the most complete analysis of candidate
designs for third-generation external-readout detectors
has been carried out by Kimble, Levin, Matsko, Thorne, kilometer-scale cavities (two arm cavities + one slosh-
and Vyatchanin [7] (KLMTV). They examined three po- ing cavity + two filter cavities). This is shown in Fig. 4
tential designs for interferometers that could beat the below.
SQL: a squeezed-input interferometer, which makes use The speed meter described in this paper can achieve
of squeezed vacuum being injected into the dark port; a a performance significantly better than a conventional
variational-output scheme in which frequency-dependent position meter, as shown in Fig. 1. (By “conven-
homodyne detection was used; and a squeezed-variational tional,” we mean “without any QND techniques.” An
interferometer that combines the features of both. (Be- example is LIGO-I.) The squeezed-input speed meter
cause the KLMTV designs measure the relative posi- (SISM) noise curve
√ shown in that Fig. 1 beats the SQL
tions of the test masses, we shall refer to them as po- by a factor of 10 in amplitude and has fixed-angle
sition meters, particularly when we want to distinguish squeezed vacuum injected into the dark port [this al-
them from the speed meters that, for example, use lows the interferometer to operate at a lower circulat-
variational-output techniques.) Although at least some ing power than would otherwise be necessary to achieve
of the KLMTV position-meter designs have remarkable that level of sensitivity, as described by Eq. (3) be-
performance in the lossless limit, all of them are highly low]. The squeezed-variational position meter (SVPM),
susceptible to losses. which requires frequency-dependent squeezed vacuum
In addition, we note that the KLMTV position meters and homodyne detection, is more sensitive than the
each require four kilometer-scale cavities (two arm cav- squeezed-input speed meter over much of the frequency
ities + two filter cavities). The speed meters described range of interest, but the speed meter has the ad-
in this paper require at least three kilometer-scale cavi- vantage at low frequencies. It should also be noted
ties [two arm cavities + one “sloshing” cavity (described that the squeezed-variational position meter requires four
below)]. If we use a variational-output technique, as kilometer-scale cavities (as described in the previous
KLMTV did, the resulting interferometer will have five paragraph), whereas the squeezed-input speed meter re-

2
quires three.
If frequency-dependent homodyne detection is added
to the squeezed-input speed meter, the resulting
squeezed-variational speed meter (SVSM) can be opti-
mized to beat the squeezed-variational position meter
over the entire frequency range. Figure 1 contains two
squeezed-variational speed meter curves; one is optimized
to match the squeezed-input speed meter curve at low fre- G(η) F(ζ)
quencies, and the other is optimized for comparison with
the squeezed-variational postion-meter curve (resulting
in less sensitivity at high frequencies).
The original idea for a speed meter, as a device for
measuring the momentum of a single test mass, was con- Tp Tb
ceived by Braginsky and Khalili [12] and was further de- J(η) C(η)
veloped by Braginsky, Gorodetsky, Khalili, and Thorne
I(ζ) B(ζ)
[13] (BGKT). In their appendix, BGKT sketched a de- D(η) A(ζ)
sign for an interferometric gravity wave speed meter and
L(ζ)
speculated that it would be able to beat the SQL. This
was verified in Ref. [8] (Paper I), where it was demon- To K(η)
strated that such a device could in principle beat the Ts
Q(η) P(ζ)
SQL by an arbitrary amount over a wide range of fre-
quencies. However, the design presented in that paper, FIG. 2. Simple version of three-cavity design for
which we shall call the two-cavity speed-meter design, had speed-meter interferometer. The main laser input port is de-
three significant problems: it required (i) a high circulat- noted by I(ζ), where ζ = t − z/c. The signal is extracted at
ing power (∼ 8 MW to beat the SQL by a factor of 10 in the bottom mirror [denoted Q(η), where η = t+z/c]. The dif-
noise power at 100 Hz and below), (ii) a large amount of ference between the one- and two-port versions is the mirror
power coming out of the interferometer with the signal shown in gray.
(∼ 0.5 MW), and (iii) an exorbitantly high input laser
power (> ∼ 300 MW). The present paper describes an al- relative velocity of the test masses1 . The sloshing fre-
ternate class of speed meters that effectively eliminate quency is
the latter two problems, and techniques for reducing the
√
needed circulating power are discussed. These improve- c Ts
ments bring interferometric speed meters into the realm Ω= , (1)
2L
of practicality.
A simple version of the three-cavity speed-meter design where Ts is the power transmissivity of the sloshing mir-
to be discussed in this paper is shown in Fig. 2. In (an ror, L is the common length of all three cavities, and c is
idealized theorist’s version of) this speed meter, the in- the speed of light. We read the velocity signal [Q(η)] out
put laser light [with electric field denoted I(ζ) in Fig. 2] at a extraction mirror (with transmissivity To ), which
passes through a power-recycling mirror into a standard gives a signal-light extraction rate of
Michelson interferometer. The relative phase shifts of the
two arms are adjusted so that all of the input light re- cTo
δ= . (2)
turns to the input port, leaving the other port dark [i.e., L
the interferometer is operating in the symmetric mode so
We have used the extraction mirror to put the sloshing
D(η) = 0 in Fig. 2]. In effect, we have a resonant cavity
cavity parallel to one of the arms of the Michelson part
shaped like ⊥. When the end mirrors move, they will
of the interferometer, allowing this interferometer to fit
put a phase shift on the light, causing some light to enter
the antisymmetric mode (shaped like ⊢) and come out
the dark port. So far, this is the same as conventional
interferometer designs (but without the optical cavities 1
in the two interferometer arms). The net signal is proportional to the relative velocities of
the test masses, assuming that the frequencies ω of the test
Next, we feed the light coming out of the dark port
masses’ motion are ω ≪ Ω = (sloshing frequency). However,
[D(η)] into a sloshing cavity [labeled K(η) and L(ζ)
the optimal regime of operation for the speed meter is ω ∼ Ω.
in Fig. 2]. The light carrying the position information
As a result, the output signal contains a sum over odd time
sloshes back into the “antisymmetric cavity” with a phase derivatives of position [see the discussion in Sec. III A]. There-
shift of 180◦ , cancelling the position information in that fore, the speed meter monitors not just the relative speed of
cavity and leaving only a phase shift proportional to the the test masses, but a mixture of all odd time derivatives of
the relative positions of the test masses.

3
into the existing LIGO facilities. The presence of the ex- Nn(η)
traction mirror essentially opens two ports to our system. T
+ e
We can use both outputs, or we can add an additional
mirror to close one port (the gray mirror in Fig. 2). We
will focus on the latter case in this paper.
The sensitivity h of this interferometer, compared to
the SQL, can be expressed as2 G(η) F(ζ)
s r
SQL
h Wcirc 800 kW
∼ ≃ , (3)
hSQL e2R Wcirc e2R Wcirc
+
Ti -
where Wcirc is the power circulating in the arms, Tp Te
SQL - + + - + C(η) + Ne(η)
Wcirc ≃ 800kW(fopt /100Hz)3 is the power required to -
reach the SQL in the absence of squeezing (for the arms of I(ζ) Tb B(ζ)
- Ti
length L = 4 km and test masses with mass m = 40 kg), + Ti
and e2R is the power squeeze factor3 . With no squeezed
D(η) A(ζ)
vacuum, the squeeze factor is e2R = 1, so the circulating Te
power Wcirc must be 8 MW in √ order to beat the SQL at W(η) -
+ + - + S(ζ)
-
fopt ≃ 100 Hz by a factor of 10 in sensitivity. With
a squeeze factor of e2R = 10, we can achieve the same Te To
Ts
performance with Wcirc ≃ 800 kW, which is the same as Q(η) P(ζ)
LIGO-II is expected to be.
This performance (in the lossless limit) is the same
as that of the two-cavity ( Paper I) speed meter for
the same circulating power, but the three-cavity design FIG. 3. Schematic diagram showing the practical version of
has an overwhelming advantage in terms of required in- the three-cavity speed-meter design, which reduces the power
flowing through the beam splitter. Three additional mirrors,
put power. However, there is one significant problem
with transmissivity Ti , are placed around the beam splitter.
with this design that we must address: the uncomfort-
The “+” and “−” signs near the mirrors indicate the sign
ably large amount of laser power, equal to Wcirc , flowing
of the reflectivities in the junction conditions for each loca-
through the beam splitter. Even with the use of squeezed tion. The mirror shown in gray closes the second port of the
vacuum, this power will be too high. Fortunately, there interferometer.
is a method, based on the work of Mizuno [15], that will
let us solve this problem:
We add three mirrors into our speed meter (labeled Ti As claimed by Mizuno [15] and tested experimentally
in Fig. 3); we shall call this the practical three-cavity speed by Freise et al. [16] and Mason [17], when the trans-
meter. Two of the additional mirrors are placed in the missivity of the third mirror decreases from 1, the stor-
excited arms of the interferometer to create resonating age time of sideband fields in the arm cavity due to the
Fabry-Perot cavities in each arm (as for conventional in- presence of the internal mirrors will decrease. This phe-
terferometers such as LIGO-I). The third mirror is added nomenon is called Resonant Sideband Extraction (RSE);
between the beam splitter and the extraction mirror, in consequently, the third mirror is called the RSE mirror.
such a way that light with the carrier frequency resonates One special case, which is of great interest to us, occurs
in the subcavity formed by this mirror and the internal when the RSE mirror has the same transmissivity as the
mirrors. internal mirrors. In this case, the effect of the internal
mirrors on the gravitational-wave sidebands should be
exactly cancelled out by the RSE mirror. The three new
mirrors then have just one effect: they reduce the carrier
2
It should be noted that, as the power increases in Eq. (3), power passing through the beam splitter—and they can
the speed-meter performance becomes more narrow band. do so by a large factor.
Additional power and a re-optimization of some of the speed Indeed, we have confirmed that this is true for our
meter’s parameters are required to maintain the same band- speed meter, as long as the distances between the three
width at higher sensitivities. See Sec. III B for details. additional mirrors (the length of the “RSE cavity”) are
3
For an explanation of squeezed vacuum and squeeze factors, small (a few meters), so that the phase shifts added to
see, for example, KLMTV and references cited therein. In the slightly off-resonance sidebands by the RSE cavity
particular, their work was based on that of Caves [14] and are negligible. We can then adjust the transmissivities
Unruh [4]. Also, KLMTV state that a likely achievable value of the power-recycling mirror and of the three internal
for the squeeze factor (in the LIGO-III time frame) is e2R ≃ mirrors to reduce the amount of carrier power passing
10, so we use that value in our discussion. through the beam splitter to a more reasonable level.

4
 a lower power [see Eq. 3], thereby reducing the dominant
losses (which are dependent on the circulating power be-
cause they come from vacuum fluctations contributing
to the back-action). In this case, the losses are less than

Arm Cavity
Power-Recycling Mirror 4% in the range 25 − 150 Hz. As before, they are lower
at high frequencies, but they increase at low frequencies.
Using FD homodyne detection does not change the losses
significantly.
This paper is organized as follows: In Sec. II we give a
brief description of the mathematical method that we use
Arm Cavity to analyze the interferometer. In Sec. III A, we present
Laser
the results in the lossless case, followed in Sec. III B by
Filter Cavity II

RSE Mirror
a discussion of optimization methods. In Sec. III C, we
Sloshing Cavity discuss some of the advantages and disadvantages of this
design, including the reasons it requires a large circulat-
ing power. Then in Sec. IV, we show how the circulat-
Squeezed ing power can be reduced by injecting squeezed vacuum
Vacuum Circulator through the dark port of the interferometer and how the
use of frequency-dependent homodyne detection can im-
prove the performance at high frequencies. In Section V,
Filter Cavity I we discuss the effect of losses on our speed meter with the
Homodyne
detector various modifications made in Sec. IV, and we compare
FIG. 4. Schematic diagram showing the practical our interferometer configurations with those of KLMTV.
three-cavity speed-meter design with squeezed vacuum in- Finally, we summarize our results in Sec. VI.
jected at the dark port and two filter cavities on the output.
Note that the circulator is a four-port optical device that sep-
arates the injected (squeezed) input and the interferometer’s II. MATHEMATICAL DESCRIPTION OF THE
output. INTERFEROMETER

The interferometers in this paper are analyzed using

With this design, the high circulating power is con- the techniques described in Paper I (Sec. II). These meth-
fined to the Fabry-Perot arm cavities, as in conventional ods are based on the formalism developed by Caves and
LIGO designs. There is some question as to the level of Schumaker [18,19] and used by KLMTV to examine more
power that mirrors will be able to tolerate in the LIGO- conventional interferometer designs. For completeness,
III time frame. Assuming that several megawatts is not we will summarize the main points here.
acceptable, we shall show that the circulating power can The electric field propagating in each direction down
be reduced by injecting fixed-angle squeezed vacuum into each segment of the interferometer is expressed in the
the dark port, as indicated by Eq. (3). form
Going a step farther, we shall show that if, in addi-
r
tion to injected squeezed vacuum, we also use frequency- 4π~ω0
dependent (FD) homodyne detection, the sensitivity of Efield (ζ) = A(ζ) . (4)
Sc
the speed meter is dramatically improved at high fre-
quencies (above fopt ≃ 100 Hz); this is shown in Fig. 1. Here A(ζ) is the amplitude (which is denoted by other
The disadvantage of this is that FD homodyne detection letters—B(ζ), P (ζ), etc.—in other parts of the interfer-
requires two filter cavities of the same length as the arm ometer; see Fig. 2), ζ = t − z/c, ω0 is the carrier fre-
cavities (4 km for LIGO), as shown in Fig. 4. quency, ~ is the reduced Planck’s constant, and S is the
Our analysis of the losses in these scenarios indicates effective cross-sectional area of the light beam; see Eq. (8)
that our speed meters with squeezed vacuum and/or of KLMTV. For light propagating in the negative z direc-
variational-output are much less sensistive to losses than tion, ζ = t−z/c is replaced by η = t+z/c. We decompose
a position meter using those techniques (as analyzed by the amplitude into cosine and sine quadratures,
KLMTV). Losses for the various speed meters we dis-
cuss here are generally quite low and are due primar- A(ζ) = A1 (ζ) cos ω0 ζ + A2 (ζ) sin ω0 ζ , (5)
ily to the losses in the optical elements (as opposed to
mode-mismatching effects). Without squeezed vacuum, where the subscript 1 always refers to the cosine quadra-
the losses in sensitivity are less than 10% in the range ture, and 2 to sine. Both arms and the sloshing cavity
50 − 105 Hz, lower at higher frequencies, but higher at have length L = 4 km, whereas all of the other lengths zi
low frequencies. Injecting fixed-angle squeezed vacuum are short compared to L. We choose the cavity lengths
into the dark port allows this speed meter to operate at to be exact half multiples of the carrier wavelength so

5
ei2ω0 L/c = 1 and ei2ω0 zi /c = 1. There will be phase shifts (vacuum-fluctuation) noise coming in the main laser port
put onto the sideband light in all of these cavities, but (ĩ1,2 ) since that noise largely exits back toward the laser
only the phase shifts due to the long cavities are signifi- and produces negligible noise on the signal light exit-
cant. ing the output port. As a result of these assumptions,
The aforementioned sidebands are put onto the carrier the only (vacuum-fluctuation) noise that remains is that
by the mirror motions and by vacuum fluctuations. We which comes in through the output port (p̃1,2 ). An in-
express the quadrature amplitudes for the carrier plus terferometer in which this is the case and in which light
the side bands in the form absorption and scattering are unimportant (R + T = 1
Z ∞ for all mirrors, as we have assumed) is said to be “loss-
  dω
Aj (ζ) = Aj (ζ) + ãj (ω)e−iωζ + ã†j (ω)eiωζ . (6) less.” In Sec. V, we shall relax these assumptions; i.e., we
0 2π shall consider lossy interferometers.
It should be noted that the results and discussion in
Here Aj (ζ) is the carrier amplitude, ãj (ω) is the field this section and in Sec. IV apply to both the simple and
amplitude (a quantum mechanical operator) for the side- practical versions of the three-cavity speed meter (Figs. 2
band at sideband frequency ω (absolute frequency ω0 ±ω) and 3). The two versions are completely equivalent (in
in the j quadrature, and ã†j (ω) is the Hermitian adjoint of the lossless limit).
ãj (ω); cf. Eqs. (6)–(8) of KLMTV, where commutation
relations and the connection to creation and annihilation
operators are discussed. In other portions of the interfer- A. Mathematical Analysis
ometer (Fig. 2), Aj (ζ) is replaced by, e.g., Cj (ζ); Aj (ζ),
by Cj (ζ); ãj (ω), by c̃j (ω), etc. The lossless interferometer output for the speed meters
Since each mirror has a power transmissivity and com- in Fig. 2 and 3, as derived by the analysis sketched in the
plementary reflectivity satisfying the equation T +R = 1, previous section, is then
we can write out the junction conditions for each mirror
in the system, for both the carrier quadratures and the L∗ (ω)
q̃1 = − p̃1 , (8a)
sidebands [see particularly Eqs. (5) and (12)–(14) in Pa- L(ω)
per I]. We shall denote the power transmissivities for the √
2iω ω0 δWcirc L∗ (ω)
sloshing mirror as Ts , for the extraction (output) mirror q̃2 = √ x̃ − p̃2 . (8b)
as To , the power-recycling mirror as Tp , for the beam- ~cLL(ω) L(ω)
splitter as Tb = 0.5, for the internal mirrors as Ti , and Here p̃j (ω) is the side-band field operator [analog of
for the end mirrors as Te ; see Figs. 2 and 3. ãj (ω) in Eq. (6)] associated with the dark-port input
The resulting equations can be solved simultaneously P (ζ), and q̃j (ω) associated with the output Q(η); see
to get expressions for the carrier and sidebands in each Fig. 2. Also, in Eqs. (8), L(ω) is a c-number given by
segment of the interferometer. Since those expressions
may be quite complicated, we use the following assump- L(ω) = Ω2 − ω 2 − iωδ (9)
tions to simplify our results. First, we assume that only √
the cosine quadrature is being driven (so that the carrier [recalling that Ω = c Ts /2L is the sloshing frequency,
sine quadrature terms are all zero). Second, we assume δ = cTo /L the extraction rate], the asterisk in L∗ (ω)
that the transmissivities obey denotes the complex conjugate, x̃(ω) is the Fourier
transform of the relative displacement of the four test
1 ≫ To ≫ Ts ≫ Te and 1 ≫ {Tp , Ti } ≫ Te . (7) masses—i.e., the Fourier transform of the difference in
lengths of the interferometer’s two arm cavities—and
The motivations for these assumptions are that (i) they Wcirc is the circulating power in the each of the inter-
lead to speed-meter behavior; (ii) as with any interfer- ferometer’s two arms. Note that the circulating power
ometer, the best performance is achieved by making the (derived as in Sec. II B of Paper I) is related to the car-
end-mirror transmissivities Te as small as possible; and rier amplitude B1 in the arms by4
(iii) good performance requires a light extraction rate
comparable to the sloshing rate, δ ∼ Ω [cf. the first para- 1 4~ω0 I12
Wcirc = ~ω0 B12 = , (10)
graph of Sec. III B
√ in Paper I], which with Eqs. (1) and 2 Ti Tp
(2) implies To ∼ Ts so To ≫ Ts . Throughout the pa-
per, we will be using these assumptions, together with
ωL/c ≪ 1, to simplify our expressions.
4
Equation (10) refers specifically to the practical version of
the three-arm interferometer (Fig. 3). The simple (Fig. 2)
III. SPEED METER IN THE LOSSLESS LIMIT version would be
1 ~ω0 I12
For simplicity, in this section we will set Te = 0 (end Wcirc = ~ω0 B12 = .
2 Tp
mirrors perfectly reflecting). We will also neglect the

6
where I1 is the input laser amplitude (in the cosine
quadrature). Readers who wish to derive the input–
output relations (8) for themselves may find useful guid- δ = √2 Ω
ance in Appendix B of KLMTV [7] and in Secs. II and III
of Paper I [8], which give detailed derivations for other
interferometer designs. δ = 0.5Ω

κ(ω)
Notice that the first term in Eq. (8b) contains x̃ only
in the form ω x̃; this is the velocity signal [actually, the
sum of the velocity and higher odd time derivatives of δ = 0.1Ω
position because of the L(ω) in the denominator]. The
test masses’ relative displacement x̃(ω) is given by
√
8i ~ω0 δWcirc 0.1 0.2 0.5 1 2 5 10
x̃ = x̃e − x̃n = Lh̃ − √ p̃1 , (11) ω/Ω
mω cLL(ω) FIG. 5. The coupling constant κ(ω) in arbitrary (logrith-
mic) units with ω measured in units of Ω. The three curves
where x̃e is the Fourier transform of the relative displace- correspond to the same light power (such that κmax√= 5 for
ment of the mirrors of the “east” arm and x̃n is the same the middle curve), but δ = 0.1Ω, δ = 0.5Ω, and δ = 2Ω.
for the “north” arm. The last term is the back-action
produced by fluctuating radiation pressure (derived as in
Sec. II B of Paper I). (radiation-pressure) noise. This might not be obvious
It is possible to express Eqs. (8) in a more concise form, from Eqs. (12), especially because they have an identical
similar to Eqs. (16) in KLMTV: form (except for the frequency dependence of κ) as the
input-output relations of a conventional interferometer,
q̃1 = ∆p̃1 = p̃1 e2iψ , (12a) where the term proportional to K (their version of κ) is
√ h̃ iψ the radiation-pressure noise. Indeed, if one measures the
q̃2 = ∆p̃2 + 2κ e , ∆p̃2 = (p̃2 − κp̃1 )e2iψ . (12b) “sine” quadrature of the output, q̃2 , as is done in a con-
hSQL
ventional interferometer, this speed meter turns out to
Here be SQL limited, as are conventional interferometers.
Fortunately, the fact that κ is constant (and equal to
Ω2 − ω 2 κ 0 ) over a broad frequency band will allow the afore-
tan ψ = − (13)
ωδ mentioned cancellation of the back-action, resulting in a
QND measurement of speed. To see this, suppose that,
is a phase shift put onto the light by the interferometer,
instead of measuring the output phase quadrature q̃2 , we
16ω0 δWcirc use homodyne detection to measure a generic, frequency-
κ= (14) independent quadrature of the output:
mcL|L(ω)|2
√ h
is a dimensionless coupling constant that couples the q̃Φ = ∆p̃1 cos Φ + (∆p̃2 + 2κ eiψ ) sin Φ , (16)
gravity wave signal h̃ into the output q̃2 , and hSQL
r
8~ where Φ is a fixed homodyne angle. Then from Eqs. (8),
hSQL = (15) we infer that the noise in the signal, expressed in GW
mω 2 L2
strain units h, is
is the standard quantum limit for a conventional inter-
ferometer such as LIGO-I or VIRGO [1]. hSQL
hn = √ eiψ [p̃1 (cot Φ − κ) + p̃2 ] . (17)
In Fig. 5, we plot the coupling constant κ as a function 2κ
of frequency for several values of δ. As the graph shows,
κ can be roughly constant for a rather broad frequency By making cot Φ = κ0 ≡ (constant value of κ at ω < ∼ Ω),
band ω ∼ < Ω, when δ is chosen to be ∼ Ω (as it will be the radiation pressure noise in hn will be cancelled in the
when the interferometer is optimized). Combining this broad band where κ = κ0 , thereby making this a QND
with the fact that hSQL ∝ 1/ω, we infer from Eqs. (12) interferometer.
that the output signal at frequencies ω < We assume for now that ordinary vacuum enters the
∼ Ω is propor-
tional to ω h̃, or equivalently ω x̃, which is the relative output port of the interferometer; i.e., p̃1 and p̃2 are
quadrature amplitudes for ordinary vacuum (we will in-
speed of the test masses (as mentioned above).
The terms ∆p̃1 and ∆p̃2 in Eqs. (8) represent quan- ject squeezed vacuum in Sec. IV A). This means [Eq. (26)
of KLMTV] that their (single-sided) spectral densities
tum noise (shot noise, radiation-pressure noise, and
their correlations). We shall demonstrate below that, are unity and their cross-correlations are zero, which,
when combined with Eq. (17), implies a spectral density
in the frequency band ω < ∼ Ω where the interferom-
eter samples only the speed, there is no back-action of

7
 Shn = (hSQL )2 ξ 2 . (18) 100
50
Here
(cot Φ − κ)2 + 1 δ = 0.5ωopt
10
ξ2 ≡ (19)
2κ 5

κ(ω)
δ = 2ωopt
is the fractional amount by which the SQL is beaten (in
units of squared amplitude). This expression for ξ 2 is the 1
same as that for the speed meters in Paper I [Eq. (35)] δ = 4ωopt
0.5
and BGKT [Eq. (40)], indicating the theoretical equiva-
lency of these designs. In those papers, an optimization
is given for the interferometer. Instead of just using the
results of that optimization, we shall carry out a more 0.1 0.2 0.5 1 2 5 10
ω/ωopt
comprehensive study of it5 . FIG. 6. The coupling constant κ(ω) with ω measured in
units of ωopt . The solid curve is determined by setting
δ = 2ωopt and κmax = 5 (this value of κmax comes from spec-
B. Optimization ifying that we want to beat the SQL by a factor of 10; see
Fig. 7). If, in addition, we set ωopt = 2π × 100 Hz, then all
The possible choices of speed meter parameters can be the parameters have been specified (due to the various rela-
investigated intuitively by examining the behavior of κ. tionships between them) and are equal to the values given in
To aid us in our exploration, we choose (as in BGKT and Table I. If we maintain the same power but change δ, then
Paper I) to express |L(ω)|2 [Eq. (9)] as the only parameter of Table I that is affected is To . Examples
of such a change are shown for δ = 0.5ωopt and δ = 4ωopt .
|L(ω)|2 = (ω 2 − ωopt
2
)2 + δ 2 (ωopt
2
+ δ 2 /4) , (20) Note that these two choices of δ are more extreme than would
be desirable in practice, but they are shown here to illustrate
where more clearly the effect on κ of changing the ratio between δ
p and ωopt .
ωopt = Ω2 − δ 2 /2 , (21)

is the interferometer’s “optimal frequency,” i.e., the fre- Ω3I

quency at which |L(ω)| reaches its minimum. Combining κmax = 2 . (24)
δ(ωopt+ δ 2 /4)
with Eq. (14), we obtain
By setting
Ω3I δ
κ= 2 2 2 + δ 2 /4) , (22) cot Φ = κmax , (25)
(ω − ωopt )2 + δ 2 (ωopt

where we get the maximum amount by which a speed meter can

beat the SQL
16ω0 Wcirc
Ω3I ≡ (23) 2
mLc 2 1 δ(ωopt + δ 2 /4)
ξmin = = . (26)
is a frequency scale related to the circulating power. At 2κmax 2Ω3I
ωopt , κ reaches its maximum (see Fig. 6) As ω differs from ωopt in either direction, κ decreases
from κmax . This causes the noise to increase since (i)
the term (cot Φ − κ)2 in the numerator of ξ 2 [Eq. (19)]
increases and (ii) the denominator of ξ 2 decreases. In
5
It should be noted that the expressions given in Sec. III A order to have broadband performance, we should make
are accurate to 6% or better over the frequency range of inter- the peak of κ(ω) as flat as possible. As we can see from
est. To achieve 1% accuracy, we expand to the next-highest both Eq. (22) and Fig. 6, the shape of the peak can be
order. The result can be expressed as a re-definition of the adjusted by changing δ: for the same optical power, a
sloshing frequency larger δ means a wider peak but a smaller maximum.
Ω2 → Ω′2 = Ω2 − δδs /2 , Therefore, changing δ is one method of balancing sensi-
tivity against bandwidth. Some examples are shown in
where δs = cTs /2L. Then κ retains the same functional form: Figs. 6, 7, and 8, where κ(ω), ξ 2 (ω), and Sh (ω), respec-
tively, are plotted for configurations with the same ωopt
16ω0 δWcirc
κ → κ′ = . and optical power Wcirc , but with several values of δ.
mcL((Ω′2 − ω 2 ) + ω 2 δ 2 )
To be more quantitative, a simple analytic form for
As a result, the optimization described in Sec. III B applies ξ 2 (ω) can be obtained by inserting Eqs. (22), (24), and
equally well to κ′ and Ω′ as to the original κ and Ω. (26) into Eq. (19) to get

8
 1 as is done by BGKT. For ξmin 2
= 0.1, this gives δ =
1.977ωopt ≈ 2ωopt (as assumed in BGKT and Paper I).
0.8 Plugging these numbers into Eq. (26) and combining with
Eq. (23) gives
0.6 3
2
mLc ωopt
ξ Wcirc (δ = 2ωopt ) = 2
8ω0 ξmin
0.4 δ = 1.5ωopt  3  
t
ω op ωopt m
2.5 ≃ 8.4 MW
0.2 δ= 
2 π × 100 Hz

40 kg
 
15
δ = 2ωopt L 1.78 × 10 Hz 0.1
× 2 . (30)
4000 km ω0 ξmin
0 0.5 1 1.5 2
ω/ωopt
Therefore, when ωopt is chosen at 2π × 100 Hz, this speed
FIG. 7. The squared amount by which the speed meter meter (with δ = 2ωopt ) requires Wcirc ≃ 8.4 MW to beat
beats the SQL with a given circulating power, which is deter- the SQL by a factor of 10 in power (ξmin2
= 0.1). [Note
2
mined by setting (for the solid curve) ξmin = 0.1 and the con-
2 that, keeping δ = 2ωopt , the speed meter reaches the SQL
dition (29). Note that the requirement on ξmin sets the power SQL
SQL
relative to the SQL power Wcirc , the value of which is depen- with Wcirc = 840 kW, comparable to the value given by
dent on ωopt . (For ωopt = 100 Hz, we have Wcirc = 8 MW.) If KLMTV Eq. (132) for conventional interferometers with
we hold the power fixed and change δ to 1.5ωopt and 2.5ωopt , 40-kilogram test masses.] The ξ 2 and Sh curves for this
we get the other two curves. configuration are plotted as solid lines in Fig. 7 and 8,
respectively.
[Sh (f)/SSQL (100 Hz)]1/2

10
SQ t Please note that Eq. (30) should be applied with cau-
ω op t Φ
1.5
L 2
5 = >c
o tion because significantly changing ξmin in the above
δ x
κm a
op
t equation (without changing the ratio between δ and ωopt )
= 2ω
δ will change the wide-band performance of the interferom-
2 pt
. 5 ωo eter, since there is some “hidden” power dependence in
2
1 δ= Eq. (29). To determine the behavior of the speed me-
2
ter with significantly higher power or lower ξmin while
0.5 maintaining the same wideband performance, we must
re-apply the requirement (29) to determine the appro-
0.2 priate ratio between δ and ωopt . For example, solving
2
Eqs. (26) and (29) simultaneously for ξmin and δ, with
chosen values Wcirc = 20 MW and ωopt = 2π × 100 Hz,
0.1 0.2 0.5 1 2 5 10 −2
ω/ωopt gives δ = 2.334ωopt and ξmin ≃ 17. Keeping this in mind,
FIG. 8. Noise curves corresponding to the ξ curves in 2 a general expression for the circulating power is
Fig. 7, the caption of which describes the parameters used
2
here as well. The dotted line is an example of a noise curve mLc(ωopt + δ 2 /4)δ
Wcirc = 2
for which κ is not quite flat and cot Φ was chosen to be slightly 32ω0 ξmin
smaller than κmax (see the end of Sec. III B for details.) " #
2
209 kW (ωopt + δ 2 /4)δ
= 2
  ξmin (2π × 100 Hz)3
2 1 ∆2    
ξ (ω) = 1 + ∆ + 4 ξ2 . (27) m L 1.78 × 1015 Hz
4ξmin (1 + ∆) min × , (31)
40 kg 4000 km ω0
Here

2 where the relationship between δ and ωopt determines
ω 2 − ωopt
2 whether the noise curve is deep but narrow or wide but
∆≡ 2 2 (28) shallow [with the requirement (29) giving the latter].
δ (ωopt + δ 2 /4)
So far, we have only changed δ to modify the perfor-
is a dimensionless offset from the optimal frequency ωopt . mance of the speed meter. Another method is to change
From Eq. (28), it is evident√that ∆, and thus ξ 2 , are the ωopt . In this case, the shape of the noise curve changes
same for ω = 0 and ω = 2ωopt [see also Eq. (47) of very little, but the minima occur at different frequen-
BGKT or Eq. (49) of Paper I]. For definiteness, let us cies, causing the interferometer to have either broader
impose that bandwidth or higher sensitivity (relative to the SQL).
This is shown in Fig. 9. Maintaining condition (29)
√ 3 2 with ωopt chosen at 2π × 150 Hz, we get a broader but
ξ 2 (0) = ξ 2 ( 2ωopt ) = ξmin (29) shallower curve (short dashes); this configuration beats
2

9
 [Sh (f)/SSQL (100 Hz)]1/2
10 SQ C. Discussion of Three-Cavity Speed-Meter Design
L H z
5 = 75
f opt z In this section, we discuss how the three-cavity speed-
0 0H
=1 meter design compares to the two-cavity design presented
2 f opt Hz in Paper I, focusing on the three major problems of that
0
15 design: it required (i) a high circulating power, (ii) a
1 t
=
f op large amount of power coming out of the interferometer
0.5 with the signal, and (iii) an exorbitantly high input laser
power.
0.2 With the three-cavity speed meter, we are able to repli-
cate the performance of the two-cavity design in Paper I,
10 20 50 100 200 500 1000 2000
but without the exorbitantly high input power. The rea-
f (Hz) son why our three-cavity speed meter does not need a
FIG. 9. Noise curves for varying optimal frequencies. The high input power is the same as for conventional inter-
solid curve has fopt = 100 Hz and is identical to the solid ferometers: in both cases, the excited cavities are fed
curve of Fig. 8. Maintaining the same power and the condition directly by the laser. According to Bose statistics, car-
imposed by Eq. (29), we show two examples of noise curves rier photons will be “sucked” into the cavities, produc-
with other optimal frequencies, specifically fopt = 75 Hz and ing a strong amplification. This was not the case in the
fopt = 150 Hz. two-cavity speed meter of Paper I. There, an essentially
empty cavity stood between the input and the excited
TABLE I. Three-arm speed-meter interferometer parame- cavity, thereby thwarting Bose statistics and resulting
ters and their fiducial values, as used throughout except where in a required input laser power much greater than the
other parameters are specified. power that was circulating in the excited cavity (see Pa-
Parameter Symbol Fiducial Value per I for more details). In this paper, we have returned
carrier frequency ω0 1.78 × 1015 s−1 to a case where the laser is driving an excited cavity di-
mirror mass m 40 kg rectly, thereby allowing the input laser power to be small
arm length L 4 km relative to the circulating power.
sloshing mirror transmissivity Ts 0.0008 Because the cavity from which we are reading out the
output mirror transmissivity To 0.017 signal does not contain large amounts of carrier light (by
end mirror transmissivity Te 2 × 10−5 contrast with the two-cavity design), this three-cavity
internal and RSE mirror trans. Ti 0.005 speed meter does not have large amounts of power ex-
optimal frequency ωopt 2π × 100 Hz iting the interferometer with the velocity signal, unlike
sloshing frequency Ω 2π × 170 Hz the two-cavity design. By making use of the different
extraction rate (half-bandwidth) δ 2π × 200 Hz modes of the Michelson interferometer, we have solved
SQL
SQL circulating power Wcirc 820 kW the problem of the exorbitantly high input power and
the problem of the amount of light that comes out of the
interferometer.
−2 The problem of the high circulating power Wcirc , un-
the SQL by a factor of ξmin ∼ 4.7, up to f ∼ 240 Hz. fortunately, is not solved by the three-cavity design. This
With ωopt = 2π × 75 Hz, we get a narrower but deeper is actually a common characteristic of “external-readout”
curve (long dashes), which beats the SQL by a factor of interferometer designs capable of beating the SQL. The
−2
ξmin ∼ 17, up to f ∼ 100 Hz. The power was kept fixed reason for this high power is the energetic quantum limit
at Wcirc = 8.2 MW. (EQL), which was first derived for gravitational-wave
One more potential optimization method is to choose interferometers by Braginsky, Gorodetsky, Khalili and
a κ with a peak that is not quite flat and then choose a Thorne [20]. The EQL arises from the phase-energy un-
cot Φ that is slightly smaller than κmax . This will give certainty principle
a wider bandwidth on either side of ωopt , at the price of
decreased sensitivity at the region near ωopt (see dotted ~ω0
line in Fig. 8). ∆E∆φ ≥ , (32)
2
For simplicity, we will choose a typical (but somewhat
arbitrary) set of parameters for the lossless interferom- where E is the stored energy in the interferometer and
eter of Fig. 2. These values, given in Table I, will be φ is the phase of the light. The uncertainty ∆E of the
used (except as otherwise noted) for subsequent plots and stored light energy during the measurement process must
calculations comparing this speed-meter design to other be large enough to allow a small uncertainty ∆φ in the
configurations. stored light’s optical phase, in which the GW signal is
contained. pFor an interferometer with coherent light (so
∆E = ~ω0 E/~ω0 ), the EQL dictates that the energy
stored in the arms must be larger than

10
 mL2 ω 2 ∆ω A. Injection of Squeezed Vacuum into Dark Port
Eξ ∼ (33)
4ω0 ξ 2
Because the amount of circulating power required by
in order to beat the SQL by a factor of ξ near frequency
our speed meter remains uncomfortably large, it is de-
ω with a bandwidth ∆ω (Eq. (1) of Ref. [11] and Eq. (29)
sirable to reduce it by injecting squeezed vacuum into
of Ref. [20]). In a broadband configuration with ∆ω ∼ ω,
the dark port. The idea of using squeezed light in
we have
gravitational-wave interferometers was first conceived by
mL2 ω 3 Caves [14] and further developed by Unruh [4] and
Eξ ∼ . (34) KLMTV. We shall start in this section with a straightfor-
4ω0 ξ 2
ward scheme that will decrease the effective circulating
For comparison, in the broadband regime of the speed power without otherwise changing the speed meter per-
meter, we have, from Eq. (26), formance.
As discussed in Sec. IV B and Appendix A of KLMTV,
2
mL2 δ(ωopt
2
+ δ 2 /4) mL2 ωopt
3
a squeezed input state is related to the vacuum input
ξmin = ∼ , (35) state (assumed in Sec. III A) by a unitary squeeze oper-
4 Eω0 4 Eω0
ator S(R, λ) [see Eqs. (41) and (A5) of KLMTV]
where the stored energy is E = 2Wcirc L/c. Comparison
between Eqs. (34) and (35) confirms that our speed meter |ini = S(R, λ)|0i . (36)
is EQL limited.
Here R is the squeeze amplitude and λ is the squeeze an-
As a consequence of the EQL, designs with coherent
gle, both of which in principle can depend on sideband
light will all require a similarly high circulating power
frequency. However, the squeezed light generated using
in order to achieve a similar sensitivity. Moreover, given
nonlinear crystals [21,22] has frequency-independent R
the sharp dependence E ∝ ω 3 , this circulating power
and λ in our frequency band of interest, i.e., f < 10 kHz
problem will become much more severe when one wants
[23]; and in this section, we shall assume frequency inde-
to improve sensitivities at high frequencies.
pendence.
Nevertheless, the EQL in the form (33) above only
The effect of input squeezing is most easily understood
applies to coherent light. Using nonclassical light will
in terms of the following unitary transformation,
enable the interferometer to circumvent it substan-
tially. One possible method was invented by Braginsky, |ini → S † (R, λ)|ini = |0i (37a)
Gorodetsky, and Khalili [10] using a special optical topol-
ogy and intracavity signal extraction. A more conven- p̃j → S † (R, λ)p̃j S(R, λ) , (37b)
tional solution for our external-readout interferometer is q̃j → S † (R, λ)q̃j S(R, λ) , (37c)
to inject squeezed light into the dark port, as we shall
discuss in Sec. IV A (and as was also discussed in the where j = 1, 2. This brings the input state back to vac-
original paper [20] on the EQL). uum and transforms the input quadratures into linear
combinations of themselves, in a rotate-squeeze-rotate
way [Eq. (A8) of KLMTV, in matrix form]:
IV. SQUEEZED VACUUM AND FD HOMODYNE      
DECTECTION p̃1 p̃1s † p̃1
→ = S (R, λ) S(R, λ)
p̃2 p̃2s p̃2
   −R 
In this section, we discuss two modifications to the cos λ − sin λ e 0
=
three-cavity speed-meter design analyzed in Sec. III A. sin λ cos λ 0 eR
This discussion applies to both the simple and practi-   
cos λ sin λ p̃1
cal versions, shown in Figs. 2 and 3; the modifications × . (38)
− sin λ cos λ p̃2
are shown in Fig. 4. The first modification is to inject
squeezed vacuum (with fixed squeeze angle) into the out- In particular, the GW noise can be calculated by using
put port of the speed meter, as shown in Fig. 4. This will the squeezed noise operator [Eq. (29) of KLMTV]
reduce the amount of power circulating in the interferom-
eter. The second modification, also shown in Fig. 4, is the hns = S † (R, λ)hn S(R, λ) , (39)
introduction of two filter cavities on the output, which
allow us to perform frequency-dependent homodyne de- and the vacuum state.
tection (described in KLMTV) that will dramatically im- A special case—the case that we want—occurs when
prove the performance of the speed meter at frequencies R = constant and λ = π/2. Then there is no rotation be-
f> tween the quadratures but only a frequency-independent
∼ fopt . squeezing or stretching,

p̃1 → p̃1s = eR p̃1 , (40a)

p̃2 → p̃2s = e−R p̃2 . (40b)

11
Consequently, Eqs. (12) for the output quadratures and homodyne detection phase Φ(ω), we have, for the
q̃1,2s = S † (R, π/2)q̃1,2 S(R, π/2) are transformed into squeezed noise operator [Eqs. (39) and (38)],
q̃1s = eR p̃1 e2iψ (41a) hSQL p
  hns = − √ eiψ 1 + κ̃2

√ h̃ iψ κ
q̃2s = e−R p̃2 − κe2R p̃1 e2iψ + κe2R e . (41b) 
  
hSQL
× p̃1 cosh R cos Ψ̃ − sinh R cos Ψ̃ − 2(Ψ̃ + λ)
The corresponding noise can be put into the same form 
as Eq. (17),   
−p̃2 cosh R sin Ψ̃ − sinh R sin Ψ̃ − 2(Ψ̃ + λ) , (44)
hSQL iψ
hns = √ e [p̃1 (cot Φeff − κeff ) + p̃2 ] , (42)
κeff where
with cot Ψ̃ ≡ κ̃ ≡ κ − cot Φ . (45)
2R 2R
cot Φeff ≡ e cot Φ , κeff ≡ e κ. (43)
The corresponding noise spectral density [computed by
Since κ is proportional to the circulating power [see using the ordinary vacuum spectral densities, Sp̃1 =
Eqs. (14)], gaining a factor e2R in κ is equivalent to gain- Sp̃2 = 1 and Sp̃1 p̃2 = 0, in Eq. (44)] is
ing this factor in Wcirc .
In other words, by injecting squeezed vacuum with (hSQL )2
Sh = (1 + κ̃2 )
squeeze factor e2R and squeeze angle λ = π/2 into κ  
the interferometer’s dark port, we can achieve precisely
the same interferometer performance as in Sec. III A, × e−2R + sinh 2R[1 − cos 2(Ψ̃ + λ)] . (46)
but with a circulating light power that is lower by
Wcirc,SISM = e−2R Wcirc,OSM . (Here “SISM” means Note that these expressions are analogous to KLMTV
“squeezed-input speed meter” and “OSM” means “or- Eqs. (69)–(71) for a squeezed-variational interferometer
dinary speed meter.” Since squeeze factors e−2R ∼ 0.1 (but the frequency dependence of their K is different from
are likely to be available in the time frame of LIGO-III that for our κ). From Eq. (46), Sh can be no smaller than
[7], this squeezed-input speed meter can function with the case when
Wcirc,SISM ≃ 0.1Wcirc,OSM.
κ̃ = 0 , cos 2(Ψ̃ + λ) = 1 . (47)

B. Frequency-Dependent Homodyne Detection

The optimization conditions (47) are satisfied when

cot Φ = κ , λ = π/2 , (48)

One can take further advantage of squeezed light by
using frequency-dependent (FD) homodyne detection at which corresponds to frequency-dependent homodyne de-
the interferometer output [24–28]. As KLMTV have tection on the (frequency-independent) squeezed-input
shown, FD homodyne detection can be achieved by send- speed meter discussed in the previous section.
ing the output light through one or more optical filters (as As it turns out, the condition cot Φ = κ can read-
in Fig. 4) and then performing ordinary homodyne detec- ily be achieved by the family of two-cavity optical filters
tion. If its implemention is feasible, FD homodyne detec- invented by KLMTV and discussed in their Sec. V and
tion will dramatically improve the speed meter’s sensitiv- Appendix C. We summarize and generalize their main re-
ity at high frequencies (above fopt = 100 Hz). Note that sults in our Appendix A. The two filter cavities are both
the KLMTV design that used FD homodyne detection Fabry-Perot cavities with (ideally) only one transmitting
was called a “variational-output” interferometer; conse- mirror. They are characterized by their bandwidths, δJ ,
quently, we shall use the term “variational-output speed (where J = I, II denote the two cavities) and by their res-
meter” to refer to our speed meter with FD homodyne onant frequencies, ω0 + ξJ δJ (the ones nearest ω0 ). The
detection. Continuing the analogy, when we have both output light from the squeezed-input speed meter is sent
squeezed-input and FD homodyne detection, we will use through the two filters, and then a homodyne detection
the term “squeezed-variational speed meter.” The follow- with frequency-independent phase θ is performed on it.
ing discussion is analogous to Secs. IV and V of KLMTV. For the squeezed-variational speed meter (shown in
2
For a generic frequency-dependent6 squeeze angle λ(ω) Fig. 4) with the parameters in Table I, plus ξmin = 0.1,
4 4 −2R
δ = 2ωopt , Λ = 4 ωopt , and e = 0.1, we have
4
4 ωopt
6 κ= 2 4 (49)
For generality of the equations, we allow the squeeze angle (ω 2 − ωopt )2 + 8 ωopt
and the the homodyne phase both to be frequency dependent,
but the squeeze angle will be fixed (frequency independent) and the required filter and detection configuration is ξI =
later in the argument [specifically, in Eq. (48)]. 1.7355, δI = 2π × 91.57 Hz, ξII = −1.1133, δII = 2π ×

12
 [Sh (f)/SSQL (100 Hz)]1/2
10 SQ splitter, RSE mirror;
L
5 gle • due to mode-mismatching7 ; and
- an
ed
fix • due to the imperfect matching of the transmissivi-
2
t ties of the RSE and internal mirrors8 .
d en
1 en
ep Since the optical losses will dominate, we focus only on
cy-d
0.5 en that type of loss here. The loss at each optical element
qu will decrease the amplitude of the sideband light (which
fre
0.2 carries the gravitational-wave information) and will si-
multaneously introduce additional vacuum fluctuations
0.1 into the optical train. Schematically, for some sideband
0.1 0.2 0.5 1 2 5 10 20 ã(ω), the loss is described by
ω/ωopt
FIG. 10. Comparison of typical noise curves for fre- p p
quency-dependent and fixed-angle homodyne detection. The ã(ω) → 1 − E(ω) ã(ω) + E(ω) ñ(ω) , (50)
FD homodyne angle Φ(ω) is that of Eqs. (48) and (49); the
fixed homodyne angle Φ is that of Eq. (43); the circulating
where E is the (power) loss coefficient, and ñ(ω) is the
power is e−2R times that of Table I; and all other parame- vacuum field entering the optical train at the loss point.
ters are identical for the two interferometers and are given in It should be noted that there are various methods of
Table I. grouping these losses together in order to simplify calcu-
lations. For example, we combine all of the losses occur-
ring in the arm (or sloshing) cavities, into one loss coeffi-
114.3 Hz, and θ = π/2. [These values are reached by cient of L ∼ 20 × 10−6 [according to KLMTV Eq. (93)].
solving Eqs. (C4) of KLMTV, or by using the simpler Then we assume that the end mirrors have transmissivity
method described in Appendix A of this paper.] The Te = 2 × 10−5 , thereby absorbing all of the arm losses
resulting performance is plotted in Fig. 10. Note the into one term [see KLMTV Eq. (B5) and preceding dis-
substantial improvement at ω > ∼ ωopt . cussion].
In the case of position-meter interferometers with op- Assuming that the noise entering at the end mirrors
tical filters (the interferometers analyzed by KLMTV), of the arm cavities is denoted ñe1,2 and ñn1,2 for the
the optical losses due to the filter cavities contribute sig- east and north arms respectively, at the end mirror of
nificantly to the noise spectral density and drastically the sloshing cavity s̃1,2 , at the port-closing mirror w e1,2 ,
reduce the ability to beat the SQL. It turns out that the and at the RSE mirror m e n1,2 and m e s1,2 [representing
squeezed-variational speed meter is less sensitive to such the losses described in the previous paragraph; see Ap-
losses, as we shall see in Sec. V. pendix B 3 for details], the output of the lossy three-
cavity speed-meter system (Fig. 3; the simplified and
practical versions are no longer equivalent, since there
V. OPTICAL LOSSES will be additional losses due to the presence of the inter-
nal and RSE mirrors) is
In order to understand the issue of optical losses in √ √
this speed meter, we shall start by addressing its internal L∗ (ω) iω δδe Ω 2δδe
q̃1 = − p̃1 + (ñe1 − ñn1 ) + s̃1
losses. These include scattering and absorption at each L(ω) L(ω) L(ω)
optical element, finite transmissivities of the end mir- √ √
Te (Ω2 − ω 2 + iωδs ) iω 2δδǫ
rors, and imperfections of the mode-matching between − e1 −
w me s1
L(ω) L(ω)
cavities. The effect of external losses (i.e., losses in the √
detection system and any filter cavities) will be discussed ω 2Lδδǫ (ω − iδi )
+ √ m
e n1 , (51a)
separately. Note that the analysis in this section includes cδi L(ω)
the internal and RSE mirrors, so it applies primarily to p ∗
√
2iω ω0 To Wcirc L∗ (ω) Ω 2δδe
the speed meter designs in Figs. 3 and 4. q̃2 = √ x̃ − p̃2 + s̃2
L ~L(ω) L(ω) L(ω)

A. Internal losses
7
In this subsection, we will consider only noise resulting According to our simple analysis in Appendix C, this ef-
from losses associated with optical elements inside the fect will be insignificant in comparison with the losses in the
optical elements, so we shall ignore it.
interferometer. These occur 8
This effect is negligibly small so we shall ignore it; see Ap-
• in the optical elements: arm cavities, sloshing cav- pendix D for details.
ity, extraction mirror, port-closing mirror, beam

13
 √ √
iω δδe Te (Ω2 − ω 2 + iωδs ) TABLE III. Fiducial values for the fractional losses occur-
+ (ñe2 − ñn2 ) − w
e2 ing in various parts of the interferometer. These losses and
L(ω) L(ω)
√ √ their values are discussed in more detail in Appendix B.
iω 2δδǫ ω 2Lδδǫ(ω − iδi )
− me s2 + √ m
e n2 , (51b) Loss source Symbol Value
L(ω) cδi L(ω) arm cavity εarm 2 × 10−5
sloshing cavity εslosh 2 × 10−5
where
extraction mirror εext 2 × 10−5
p " √ √
∗
4 2~ω0 Wcirc RSE cavity εRSE 2 × 10−5
iω 2cδ iω cδǫ
x̃ = Lh̃ − √ p̃1 + √ m
e s1 port-closing mirror εclose 2 × 10−5
mcω 2 L(ω) L L local oscillator εlo 0.001
√ √ photodiode εpd 0.001
δǫ [Ω2 − iω(δ + δi )] Ω cδe
− √ e n1 − √ s̃1
m circulator εcirc 0.001
δi L mode-mismatch into filters εmm 0.001
√ #
p iω cδe Combined loss source terms
+iω δδe w e1 − √ (ñe1 − ñn1 ) (52) arms, extraction mirror, & sloshing cavitya εAES 6 × 10−5
2L
local oscillator, photodiode, & circulator εOPC 0.003
filter cavities (with mode mismatch) εF 0.005
with
a
This loss does have some weak frequency dependence, shown
δe = cTe /2L , δs = cTs /2L , in Eq. (B8), which will cause it to increase slightly at very low
δi = cTi /4L , δǫ = cE/2L . (53) frequencies.

Note that the expression for the circulating power now

has the form may have imaginary parts. For more information, in-
cluding physical explanations of each of these terms, see
∗ 1 4~ω0 Ti Tp I12 Appendix B.
Wcirc = ~ω0 B12 = (54)
2 (Ti Tp + 4Te )2 It is simple at this point to include the losses associ-
ated with optical elements external to the interferometer.
[cf. Eq. (10)]. These include losses are associated with
Equations (51) are approximate expressions [accurate
to about 6%, as were Eqs. (8); see Footnote 5], where • the local oscillator used for homodyne detection,
the assumptions (7) regarding the relative sizes of the
• the inefficiency of the photodiode,
transmissivities were used to simplify from the exact ex-
pressions. Alternatively, they can be derived analyti- • the circulator by which the squeezed vacuum is in-
cally by keeping
√ the leading order of the small quanti- jected, and
ties ωL/c ∼ Ts ∼ To ∼ Ti , plus the various loss fac-
tors; see Sec. VI of KLMTV and Sec. IV of Paper I for • the external filter cavities used for the variational-
details of the derivations for other inteferometer designs. output scheme.
In addition to confirming the approximate formulas, such
a derivation can also clarify the origins of various noise These can be addressed in the same manner as the losses
terms and their connections to one another. inside the speed meter. We need only include two more
terms in the summation, N = OPC for the local os-
cillator, photodiode, and circulator and N = F for the
B. Internal and External Losses in Compact Form filters. Again, these terms are shown in Tables II and III
and described in more detail in Appendix B.
S R
Using these EN and EN , we can rewrite the input-
In order to simplify the above Eqs. (51) and (52), we
output relation (51) in the same form as Eq. (12) as
define κ∗ in identically the same way as we defined κ
∗ S follows:
[Eq. (14) or (22)] but with Wcirc → Wcirc . Let EN and
R     
EN represent the shot and radiation-pressure noises for q̃1 1 0 p̃1
the various parts of the interferometer, specified by N . = e2iψ
q̃2 −κ∗ 1 p̃2
S R   
In Table II, expressions for EN and EN are given for N = X S
EN 0 nN 1
AES (arm cavities, extraction mirror, and sloshing cav- + e2iαN
ity combined), close (port-closing mirror), RSEin (RSE −κ∗ ENR
ENS nN 2
N
cavity in the north direction, or going “in” to the arms),  
√ h iψ 0
and RSEout (RSE cavity in the south direction, or going + 2κ∗ e , (55)
hSQL 1
“out” of the arms). The various εN represent the charac-
teristic (and frequency-independent) fractional losses for where the αN are uninteresting phases that do not affect
each of these terms; values are given in Table III. Note the noise.
S R
that, by definition, EN are required to be real, while EN

14
 S R
TABLE II. Loss factors EN due to shot noise and EN due to radiation pressure for each type of loss source in the interfer-
ometer.
S R
Source N EN (shot noise) EN (radiation pressure noise)
arm cavities, r r
εAES ωδ eiψ εAES
extract. mirror, AES −
To |L(ω)| 2 To
sloshing cavity

port-closing √ Ω2 − ω 2 ieiψ √
close εclose − εclose
mirror |L(ω)| 2
s   r
RSE cavity εRSE Ti ω2 ωδ εRSE To ω(δi + δ) + iΩ2
RSEin 1+ eiψ−iβi
“in” to arms 4To δi2 |L(ω)| Ti ωδ
s   r
RSE cavity εRSE Ti ω2 ωδ εRSE To ω(δi − δ) − iΩ2
RSEout 1+ eiψ+iβi
“out” to slosh 4To δi2 |L(ω)| Ti ωδ

local oscillator,
√
photodiode, OPC εOPC 0
and circulator
√
filter cavities F εF 0

output at homodyne angle Φ is measured, giving

21
| | AE ( 
RS S (hSQL )2 ∗ 2
0.1
RS n
Ei Shn (ω) = (cot Φ − κ ) + 1
Eo 2κ∗
)
0.01 ut F
X
S R ∗ 2 S 2
10-3 AES
OPC + |EN cot Φ − EN κ | + (EN ) , (56)
N
RSEin, out
10-4 where we have assumed all of the vacuum fluctuation
close
10-5 spectral densities are unity and the cross-correlations are
close zero; this is the same technique that we used to derive
10-6 Eqs. (18) and (46) and that was used in Paper I and
5 10 50 100 500 1000 S
f (Hz) KLMTV. Given the complicated behaviors of EN and
R
FIG. 11. Moduli-squared of the loss factors shown in Table EN , including these loss terms in the optimization of the
II. In general, the black curves are the radiation-pressure homodyne phase Φ(ω) is unlikely to be helpful. There-
noise and the gray curves are the shot noise. The parameters fore, we will use cot Φ = κ∗max , as in the lossless case.
used for this plot are given in Tables I and III. This gives us a total noise with losses:
( 
(hSQL )2 ∗ ∗ 2
Shn (ω) = (κmax − κ ) + 1
The relative magnitudes of the loss terms are shown 2κ∗
in Fig. 11. From the plot, we can see that there are sev-
X )
eral loss terms—specifically, the shot noise from the AES, + S ∗ R ∗ 2 S 2
|EN κmax − EN κ | + (EN ) . (57)
OPC, and filter cavities (if any)—that are of compara- N
ble magnitude at high frequencies and dominate there.
The AES radiation-pressure term dominates at low fre- When we inject squeezed vacuum into the dark port,
quencies, and the RSE radiation-pressure terms are also we get output operators
significant. Since the largest noise sources at low frequen-     R 
q̃1s 1 0 e p̃1
cies are radiation-pressure terms, they will be dependent = e2iψ
q̃2s −κ∗ 1 e−R p̃2
on the circulating power. Consequently, those terms will X  S
 
become smaller when the circulating power is reduced, 2iαN EN 0 nN 1
+ e
as when squeezed vacuum is injected into the dark port. −κ∗ ENR S
EN nN 2
N
This will be demonstrated in Fig. 12 below.  
√ h iψ 0
To compute the spectral noise density, we suppose the + 2κ∗ e (58)
hSQL 1

15
that can be regarded as acting on the ordinary vac- 50
OSM
uum states of the input. Once again assuming that the

[Sh (f)/SSQL(100 Hz)]1/2

vacuum fluctuation spectral densities are unity and the
cross-correlations are zero, the squeezed-input noise spec- SISM
10
tral density with homodyne detection at phase Φ is 5 SQ
(  los L
sle
(hSQL )2 ∗ 2 2R −2R ss
Shns (ω) = (cot Φ − κ ) e + e SM
2κ∗ 1
X ) SM
S R ∗ 2 S 2 0.5 SV
+ |EN cot Φ − EN κ | + (EN ) . (59)
N
5 10 50 100 500 1000
f (Hz)
FIG. 12. Noise curves showing the effects of losses.
C. Performance of Lossy Speed Meters and Noise curves for lossy versions of the ordinary (OSM),
Comparisons with Other Configurations squeezed-input (SISM), and squeezed-variational (SVSM)
speed meters are shown, along with a curve of the lossless
Examples of lossy speed meter noise curves with and ordinary speed meter for comparison. All speed meter√curves
without squeezed vacuum [Eqs. (57) and (59)] are shown here have the same parameters: δ = 2ωopt , Ω = 3ωopt ,
in Fig. 12. Note that, as mentioned before, the losses are ωopt = 2π × 100 Hz, and Ti = 0.005. The rest of the parame-
ters are given in Tables I and III.
less significant when squeezed vacuum is used to reduce
the circulating power, since the radiation-pressure noise
coming from the losses is reduced. In the ordinary p speed The broadband behaviors of the speed meters with
meter (no squeezed vacuum), the losses increase Shn by losses are particularly interesting. We start by looking
5 − 9% in the band 50 − 105 Hz. The losses have little ef- at the expression for the noise spectral density, Eq. (59).
fect above this range, but below it, noise increases signifi- An ideal (lossless) speed meter in the broadband config-
cantly, mostly due to the radiation-pressure noises shown uration beats the SQL from 0 Hz up to ω ∼ ωopt , by
in Fig. 11. For the squeezed-input speed meter p (power roughly a constant factor, because κ is roughly constant
−2R
squeeze-factor e = 0.1), the losses increase Shn by in this band. This is the essential feature of the speed
3−4% in the band 25−150 Hz. Again, the losses have lit- meter; see Sec. III. Focusing on this region, we have,
tle effect above this range. At low frequencies, however, approximately (for squeezed-input speed meters that are
the losses get quite large: 11% at 10 Hz, 32% at 5 Hz, and lossy):
73% at 3 Hz. Losses in the squeezed-variational speed
meter are much the same as in the squeezed-input speed 
h2SQL −2R X S 2
meter. The slight difference at low frequencies is due to Shns (ω) ≈ e + |EN |
2κ∗max
the fact that the lossless squeezed-variational speed me- N

ter is slightly better in that regime than the ordinary or X
squeezed-input speed meter. +κ∗ 2max S
|EN R 2
− EN | . (60)
N
The noise curves of squeezed-input speed meters (with
ordinary homodyne detection) compared with the SQL Qualitatively, we can see that if the losses are not severe
are shown in Fig. 13, along with the noise of a conven- or if κ∗max is relatively small (such that the later two terms
tional position meter with the same optical power. These in the above equation are small compared to the power
speed meters beat the SQL in a broad frequency band, squeeze factor e−2R ), the losses do not contribute signif-
despite the losses. In particular, the noise curve for the icantly to the total noise. If, in addition, the dominant
speed meter with Wcirc = 800 kW (and fopt = 107 Hz) loss factors are (almost) frequency independent, then the
matches the curve of the conventional position meter at noise due to losses gives a rather constant contribution,
high frequencies, while it beats the SQL by a factor of as shown by curves in Fig. 12. In particular, the large
∼ 8 (in power) below ∼ 150 Hz. In terms of the signal- bandwidth is preserved. (There is a slight exception to
to-noise ratio for neutron star binaries, for example, this this statement in the absence of squeezed input. Without
configuration improves upon the conventional design by squeezed input, the circulating power is higher, causing
a factor of 3.6 in signal-to-noise ratio, which corresponds κ∗max to be 10 times larger than the other cases. Conse-
to a factor of 43 increase in event rate. If it is possible quently, the frequency dependence of EAES R
to appear in
to have a higher circulating power, say Wcirc = 2 MW, the output.)
the squeezed-input speed meter would be able to beat As κmax increases, the noise from the losses may be-
the SQL by a factor of ∼ 14, corresponding to a factor of come dominant. In fact, when one minimizes the noise
4.6 in signal-to-noise and 97 in event rate. (Such a noise spectral density with respect to κ∗max , one obtains the
curve is shown in Fig. 13). following loss-dominated result:

16
 50 10
[Sh (f)/SSQL (100 Hz)]1/2 CPM ∗(ω)

8
10
5 6

SI
SM 4
1 ,8
SI 00 κ∗(ω)
0.5 SM kW
,2
MW 2
SQ
L

5 10 50 100 500 1000 0

f (Hz) 20 50 100 200 500
f (Hz)
FIG. 13. Comparison of noise curves of a conventional po- FIG. 14. Comparison of the squeezed-variational speed me-
sition meter (CPM) and squeezed-input speed meters (SISM) ter’s κ∗ with the equivalent coupling constant K∗ (as de-
with circulating powers Wcirc = 820 kW and Wcirc = 2 MW. fined by KLMTV) for the squeezed-variational position me-
The speed meters have fopt = 107 Hz, with Ω and δ de- ter. Parameters are Wcirc = 820 kW, γ = δ = 2π × 100 Hz,
termined by Eq. (29). Other parameters used are those in Ω = 2π × 173 Hz.
Tables I and III with Ti = 0.005 and e−2R = 0.1.
50
v SVPM

[Sh (f)/SSQL (100 Hz)]1/2

u ! !
u X X
Sh (ω) ≈ hSQL t
L 2 S − E R |2
|EN N e−2R + S |2 ,
|EN 10
N N
5
(61)

which is achieved if and only if 1 loss limits

SM
s 0.5 SV
P S |2
∗ L e−2R + N |EN SQ
κmax = κ ≡ P S R 2
. (62) L
N |EN − EN |
5 10 50 100 500 1000
This κL is rather constant and is comparable in mag- f (Hz)
nitude to the values of κ∗ (ω) of our speed meters, sug- FIG. 15. Comparison of noise curves for a
gesting that the speed meters can become loss-limited squeezed-variational position meter (SVPM; analyzed in
over a broad band of frequencies. Contrast this with the KLMTV) and for a squeezed-variational speed meter (SVSM;
KLMTV position meters, where K∗ (ω) grows as ω −2 at analyzed in this paper). Parameters used are those in Tables I
low frequencies; see Fig. 14. This is a fundmental prop- and III with Ti = 0.005 and e−2R = 0.1. Also shown are the
loss limits described in Sec. V C.
erty of displacement meters. As a result, a position meter
optimized at some frequency fopt may be able to reach
its “loss limit” (the equivalent of ShL ) at that frequency (where γ is the bandwidth of the arm cavities, as de-
fopt , but doing so will result in a sharp growth of noise fined in KLMTV) and if we consider high frequencies
at freqencies below fopt . In contrast, a speed meter simi- (ω >∼ {γ, Ω}). Figure 14 shows an example of this [with
larly optimized is able to stay at the noise level of its loss Wcirc = 820 kW, γ = δ = 2π × 100 Hz, Ω = 2π × 173 Hz].
limit ShL over a wide band of frequencies below fopt ; see The noise curves of the two interferometers are shown in
Fig. 15. While it is unfortunate that losses limit the per- Fig. 15.
formance of interferometers, the speed meter is at least As expected, the two noise curves in Fig. 15 agree at
able to retain a wide-band sensitivity even in the presence very high frequencies. At intermediate frequencies, the
of a loss-limit. speed meter’s κ∗ is larger than the position meter’s K∗ ,
To give a specific example of this loss-limit phe- and thus the speed meter has better sensitivity than the
nomenon, we first notice that, with the same circulat- position meter. As the frequency decreases, the speed
ing power, the conventional position-meter K∗ and our meter reaches its loss limit first and stays at that limit for
(squeezed-variational) speed-meter κ agree9 if δ = γ a wide range of frequencies. The position meter, however,
only touches its loss limit and then increases rapidly.

In fact, K∗ can be obtained from the speed meter κ∗ by

9

putting Ω → 0 and δ → γ.

17
 VI. CONCLUSIONS where
cTFP
We have described and analyzed a speed-meter inter- δFP = (A2)
4LFP
ferometer that has the same performance as the two-
cavity design analyzed in Paper I, but it does so without is the half bandwidth of the cavity. [Note that Eq. (A1) is
the substantial amount of power flowing through the sys- KLMTV Eqs. (88) and (C2), but a factor of 2 was missing
tem or the exorbitantly high input laser power required from their equations. Fortunately, this appears to be a
by the two-cavity speed meter. It was also shown that typographical error only in that particular equation; the
the injection of squeezed vacuum with e−2R = 0.1 into factor of 2 is included in their subsequent calculations.]
the dark port of the interferometer will reduce the needed As a result of this phase shift, the input (b̃1,2 )–output
circulating power by an order of magnitude, bringing it (b1,2 ) relation for sideband quadratures at frequency ω
into a range that is comparable to the expected circu- will be [KLMTV Eqs. (78)]
lating power of √LIGO-II, if one wishes to beat the SQL
by a factor of 10 in amplitude. Additional improve-    
b̃1 b1
ments to the sensitivity, particularly at high frequencies, = ei αm Rαp , (A3)
b̃2 b2
can be achieved through the use of frequency-dependent
homodyne detection. where
In addition, it was shown that this type of speed-meter
interferometer is not nearly as susceptible to losses as 1 1
αm ≡ (α+ − α− ) , αp ≡ (α+ + α− ) , (A4)
those presented in KLMTV. Its robust performance is 2 2
due, in part, to the functional form the coupling factor
κ, which is roughly constant at low frequencies. This and
helps to maintain the speed meters’ wideband perfor-  
cos φ − sin φ
mance, even in the presence of losses. Losses for the Rφ ≡ . (A5)
sin φ cos φ
various speed meters we discuss here are generally quite
low. The dominant sources of loss-induced noise at low If a frequency-independent homodyne detection at
frequencies (f < ∼ fopt ) are the radiation-pressure noise phase shift θ follows the optical filter, the measured quan-
from losses in the arm, extraction, and sloshing cavities. tity will be [KLMTV Eqs. (81) and (82)]
Because this type of noise is dependent on the circulating
power, it can be reduced by reducing the power by means b̃θ = eiαm bζ , (A6)
of squeezed input.
where

ACKNOWLEDGMENTS 1
ζ(ω) = θ − αp ≡ θ − (α+ + α− ) . (A7)
2
We thank Kip Thorne for helpful advice about its If more than one filter is applied in sequence (I, II, . . . ,)
solution and about the prose of this paper. We also and followed by homodyne detection at angle θ, the mea-
thank Farid Khalili, Stan Whitcomb, Ken Strain, and sured quadrature will be [Eq. (83)]
Phil Willems for useful discussions. This research was
supported in part by NSF grant PHY-0099568 and the 1
David and Barbara Groce Fund at the San Diego Foun- ζ(ω) = θ − (αI+ + αI− + αII+ + αII− + . . .) . (A8)
2
dation.
[Note that this ζ(ω) (KLMTV’s notation) is the same
homodyne angle Φ(ω) that we want to produce.] By
APPENDIX A: FP CAVITIES AS OPTICAL adjusting the parameters ξJ and δJ , one might be able
FILTERS to achieve the FD homodyne phases needed. KLMTV
worked out a particular case for their design [their Sec. V
As proposed by KLMTV [Sec. V B and Appendix C], B, V C, and Appendix C].
Fabry-Perot cavities can be used as optical filters to Here we shall seek a more complete solution that works
achieve frequency-dependent homodyne detection. Here in a large class of situations. With the help of Eq. (A1),
we shall briefly summarize and generalize their results. Eq. (A8) can be written in an equivalent form
Suppose we have one FP cavity of length LFP and res- Y
onant frequency ω0 − ξFP δFP . Also suppose this cavity 1 + i tan ζ 1 − i tan (αJs /2)
= e2iθ ,
has an input mirror with finite transmissivity TFP and a 1 − i tan ζ 1 + i tan (αJs /2)
J=I,II,... ,s=±
perfect end mirror. When sideband fields at frequency Y ω − s(−ξJ δJ − iδJ )
ω0 ± ω emerge from the cavity, they have a phase shift = e2iθ . (A9)
ω − s(−ξJ δJ + iδJ )
J=I,II,... ,s=±
α± ≡ 2 arctan(ξFP ± ω/δFP ) , (A1)

18
Suppose the required tan ζ(ω) is a rational function in 1. Arms, Extraction Mirror, and Sloshing Cavity
ω2, (AES)
Pn
Bk ω 2k The losses in the arms allow an unsqueezed vacuum
tan ζ(ω) = Pk=0
n 2k
, (A10) √
k=0 Ak ω field εarm ñarm to enter the optical train. By idealiz-
ing this field as arising entirely at the arm’s end mirror,
where Ak and Bk are real constants with A2n + Bn2 > 0. propagating the field through the interferometer to the
Then Eq. (A9) requires that, for all ω, output port, we obtain the following contribution to the
n
output [cf. Eq. (50)]. The associated noise can be put
X into the following form
(Ak + iBk )ω 2k
k=0 r "  
Y   εarm iψ ωδ 1 0
Narm = − e
= D eiθ ω − s(−ξJ δJ − iδJ ) , (A11) To |L(ω)| 0 1
J=I,II,... ,s=±  #  
2iψ 0 0 ñarm1
+e , (B2)
where D can be any real constant. Equation (A11) can κ∗ /2 0 ñarm2
be solved as follows. First, match the roots of the poly-
nomials of ω on the two sides of the equation; denote where the vacuum operators from the two arms are com-
these roots by ±ωJ with J = 1, 2, . . . , n. Then we can bined as
deduce that n filters are needed, and their complex reso-
nant frequencies must be offset from ω0 by ñej − ñnj
ñarm j = √ . (B3)
2
ωJ = −δJ ξJ − iδJ , J = I, II, . . . , (A12)
The first term (independent of κ∗ ) is the shot-noise con-
where ±ωI,II,... [with ℑ(ωJ ) > 0] are the 2n roots of tribution, while the second term (proportional to κ∗ ) is
the radiation-pressure noise. It turns out that several
n
X of the other loss sources N have a similar mathematical
(Ak + iBk ) ω 2k . (A13) form.
k=0
We consider, specifically, the loss from the extraction
√
After this, the polynomials on the two sides of Eq. (A11) mirror, which effectively allows εext ñext into the opti-
can only differ by a complex coefficient whose argument cal train. By propagating this field through the inter-
determines θ. In fact, by comparing the coefficients of ferometer to the output port, we obtain the following
ω 2n on both sides, we have contribution to the noise:
r "  
θ = arg(A2n + iB2n ) . (A14) εext iψ ωδ 1 0
Next = e
To |L(ω)| 0 1
 #  
2iψ 0 0 ñext1
+e . (B4)
APPENDIX B: SEMI-ANALYTICAL κ∗ /2 0 ñext2
TREATMENT OF THE LOSS TERMS
The loss from the sloshing cavity is a bit different: the
In this appendix, we present a semi-analytic treatment imperfect end mirror of the sloshing cavity produces a
√
of each source of noise included in Sec. V A. We will use vacuum noise field εslosh ñslosh which exits the cavity
a notation similar to Eq. (12), but in matrix form: with the form
    s
q̃1 q̃1 4εslosh /Ts √ iΩ
= + Nloss source , (B1) eiβs ñslosh 1,2 ≈ εslosh ñslosh 1,2 ,
q̃2 q̃2 lossless 1 + ω 2 /(δs /2)2 ω

where Nloss source is a vectorial representation of (B5)

whichever source of loss we are considering at the mo-
ment. Each of these where βs ≡ arctan(2ω/δs ) ≈ π/2 for most of the fre-
p terms is associated with a vacuum quency band of interest. The associated noise is
field of the form E(ω)ñ(ω) [cf. Eq. (50)], which enters
the interferometer and increases the level of noise present. r "  
For generality, we let E(ω) be frequency dependent. The εslosh iΩ iψ ωδ 1 0
Nslosh = − e
(constant) characteristic fractional losses for each type of To ω |L(ω)| 0 1
loss will be denoted by ε with an appropriate subscript.  #  
Each loss term appearing in Table II is presented in a 2iψ 0 0 ñslosh1
+e . (B6)
subsection below. κ∗ /2 0 ñslosh2

19
 Since the vacuum fields ñarm , ñext , and ñslosh are in- 3. The RSE Cavity
dependent and uncorrelated, we can effectively combine
these four noises into a single expression The losses in the region between the internal mirrors
r "   and the RSE mirror, i.e., the RSE cavity, are more com-
εAES iψ ωδ 1 0 plicated than the previous cases. As before, we suppose
NAES = e
To |L(ω)| 0 1 that, during each propagation from one end to the other
 #   of the RSE cavity, a fraction εRSE of the light power is
2iψ 0 0 ñAES1 dissipated and replaced by a corresponding vacuum field,
+e , (B7) √ √
κ∗ /2 0 ñAES2 εRSE ñin or εRSE ñout (depending whether the light is
propagating in towards the arms or out towards the ex-
with traction mirror and sloshing cavity). These two fields ñin
and ñout are independent vacuum fields. At the leading
εAES ∼ EAES (ω) ≡ εarm + εext + εslosh Ω2 /ω 2 . (B8) order in εRSE , we have a modified version of the “input–
output” relation for the RSE cavity:
We expect that εarm ∼ εslosh ∼ εext ∼ 2 × 10−5 , as dis- ! 
  √
cussed in the paragraph following Eq. (50) and as shown B 1 − 1+R i
ε RSE
Ri
ε RSE A
in Table III. = √ 2Ti Ti
D Ri
ε RSE 1 − 1+Ri
ε C
T 2Ti RSE
r i  √  
εRSE 1 − Ri ñin
+ √ , (B12)
2. Port-Closing Mirror Ti − Ri 1 ñout

The imperfection of the closing mirror √ has two effects: where A, B, C, D are the field amplitudes shown in Fig. 3.
(i) it introduces directly a fluctuation − εclose Ro ñclose Note that, for simplicity, we are looking at only one arm;
into the output, giving a shot noise we could equally well use the other (substituting B → F
  and C → G) or the proper combination of both. Also,
p ñclose1 notice that if εRSE = 0, then we find B = A and D = C,
Nshot
close
direct
= − ε R
close o ; (B9)
ñclose2 which illustrates the fact that the internal and RSE mir-
√ rors have no effect on the sidebands (described in Sec. I
and (ii) it introduces a fluctuation εclose To ñclose into where we introduced the RSE mirror).
the light that passes from the arms into the sloshing cav- From Eq. (B12), we find that the loss inside the RSE
ity, giving (after propagation through the sloshing cavity cavity has two effects. First, it makes the cancellation of
and interferometer and into the output): the effect of the internal and the RSE mirrors imperfect.
" (Recall that an RSE mirror with the same transmissivity
 
indirect √ iψ ωδ 1 0 as the internal mirrors effectively cancels the effect of the
Nclose = − εclose e internal mirrors on the sidebands; this was discussed in
|L(ω)| 0 1
Sec. I.) This imperfect cancellation will not be important
 #  
2iψ 0 0 ñclose1 in our situation. Indeed, there is no corresponding term
+e . (B10) appearing in the input–output relation given in Eq. (51).
κ∗ /2 0 ñclose2
Secondly, the loss inside the RSE cavity adds two vac-
Combining these two expressions gives, to leading order uum fields to light that travels through the RSE cavity
(in the various transmissivities and the small parameters in opposite directions [i.e., from A to B (IN) and from C
ωL/c and εclose ), to D (OUT)]. We denote them by
" r p
  εRSE
√ Ω2 − ω 2 1 0 ÑIN ≡ (ñin − Ri ñout ) , (B13a)
Nclose = εclose ieiψ Ti
|L(ω)| 0 1 r
εRSE p
 #   ÑOUT ≡ (− Ri ñin + ñout ) . (B13b)
2iψ 0 0 ñclose1 Ti
−e . (B11)
κ∗ /2 0 ñclose2
Note that ñin and ñout arise inside the RSE cavity as
Since εclose is simply the loss from the port-closing mirror a result of the loss that occurred there and that ÑIN
itself, we can assume that εclose < −5 and ÑOUT are the vacuum fluctuations emerging from
∼ 2 × 10 . Then, this the RSE cavity. As a result, ÑIN and ÑOUT exist in
and the above expression (B11) show that the output
noise from the closing mirror is To times smaller than different locations: ÑIN denotes the vacuum field inside
the AES loss [Eq. (B8)]. the arm cavity with B, and ÑOUT denotes the vacuum
field at the RSE mirror, heading towards the extraction
mirror and sloshing cavity with D. This is depicted in
Fig. 16.

20
 nout ÑIN and ÑOUT contribute to the carrier amplitude fluc-
tuation at the position of the test masses. Therefore, we
D C ask what the sum of ÑIN and ÑOUT is when they com-
OUT
bine at the end mirrors of the arm cavities. Since ÑOUT
IN
is superposed on D, ÑOUT must be propagated through
A B the sloshing cavity and back to the arm cavity, where it
nin is combined with ÑIN . There is a phase factor of eiωL/c
RSE Mirror ITM due to the propagation from the internal mirror to the
FIG. 16. Schematic diagram of a simplified version of the end mirror (in addition to the phases acquired on the way
RSE cavity. The quantities ñin and ñout enter inside the RSE to and inside the sloshing cavity; these are explained be-
cavity, whereas Nin and Nout are external to the cavity and
low), producing
exist in different locations.
 
e2iβs
B → B + eiωL/c ÑIN − ÑOUT (1 − To )
The fields ÑIN and ÑOUT both have a power spec- 1 − To e2iβs
r "
tral density a factor ∼ 1/Ti larger than the one-time loss εRSE ω(δi + δ) + iΩ2
coefficient. This can be explained by the fact that the ≈ B + 2To ñin
Ti ωδ
sideband light bounces back and forth inside the RSE #
cavity roughly ∼ 1/Ti times before exiting. As a result, ω(δi − δ) − iΩ2
the (power) loss coefficient is amplified by the same fac- + ñout . (B16)
ωδ
tor. However, since these fields are quite correlated (both
contain similar amounts of ñin and ñout ), we need to an- where βs = arctan(2ω/δs ) is the phase associated with
alyze them carefully. the sloshing cavity. Propagating the new B to the output
For the shot noise, we need to find the amplitude of produces a radiation-pressure contribution
the vacuum fluctuations that the loss introduces into the r  
output. To understand the effect of this type of loss, we rad pres εRSE To 2iψ 0 0
NRSE = e
ask how much vacuum fluctuation is added to the field D Ti −κ∗ 0
by ÑIN and ÑOUT . The answer is obtained by propagat- "  
ing ÑIN one round trip inside the interferometer’s arm(s) ω(δi + δ) + iΩ2 ñin1
×
and then combining it with ÑOUT . This gives ωδ ñin2
h i  #
ω(δi − δ) − iΩ2 ñout1
D → D + ÑOUT + e2iωL/c ÑIN + . (B17)
ωδ ñout2
s  
εRSE Ti ω2
≈ D+ 1+ 2 p before, this noise does not have a magnitude ∼
As
4 δi εRSE /Ti ; it is much smaller. The reason is that when
iβi −iβi


× e ñin + e ñout , (B14) ÑOUT travels to the sloshing cavity and back to the arms,
it gains two phase shifts. First is a constant phase shift
where δi ≡ Ti c/4L and βi ≡ arctan(ω/δi ). Propagating of π, due to the distance it traveled (twice) between the
this to the output, we get the shot noise contribution to RSE and sloshing mirror. The other is from the slosh-
be ing cavity, where for frequencies much larger than the
s   "   bandwidth δs of the sloshing cavity, this phase shift is
ε T ω 2 ωδ
shot RSE i iψ +iβi ñin1 roughly π. Adding these two phase shifts, ÑOUT will ap-
NRSE = 1+ 2 e e
4To δi |L(ω)| ñin2 pear roughly unchanged when it combines with ÑIN in
 # the arm cavity. Since these two vacuum fields are anticor-
ñout1 related, there is again an effective cancellation between
+e−iβi . (B15)
ñout2 the two noises at frequencies above δs . This cancellation
becomes less complete at low frequencies; see Fig. 11.
This noise is not of the magnitude that Eqs. (B13) would We assume the fractional loss εRSE ∼ 2 × 10−5 , since
appear to indicate. Instead of having a coefficient of ∼
p it arises primarily from losses in the RSE cavity’s optical
εRSE /Ti , it has a much smaller value when ω <
∼ δi . The elements (mirrors and beam splitter). (See Appendix C
reason is that the two vacuum fluctuations traveling in for a discussion of the noise due to mode mismatching,
opposite directions are anticorrelated and largely cancel which we do not consider here.)
each other, since they are summed in the outgoing field
D. This cancellation becomes less perfect as ω grows
and becomes much larger than δi . This effect is shown in 4. Detection and Filter Cavities
Fig. 11.
For the RSE contribution to the radiation-pressure First, we consider the losses involved in the detection
noise, we are interested in how much the two noise fields of the signal (without filter cavities). Two important

21
sources of photon loss are mode mismatching associated APPENDIX C: EFFECTS DUE TO
with the local oscillator used for frequency-independent MODE-MISMATCHING:
homodyne detection (εlo ) and the inefficiency of the pho- A SIMPLE ANALYSIS
todiode (εpd ). In a squeezed-input speed meter, there
will also be a circulator (with fractional loss εcirc ) through In the practical implementation of GW interferome-
which the squeezed vacuum is fed into the system and ters, the mismatching of spatial modes between differ-
through which the output light will have to pass. These ent optical cavities will degrade the sensitivity because
losses have no frequency dependence, so they are modeled signal power will be lost into higher-order modes and,
by an equation of the form of [Eq. (50)] with correspondingly, vacuum noises from those modes will
EOPC (ω) = εOPC = εlo + εpd + εcirc (B18) be introduced to the signal. In a way, this is similar to
other sources of optical loss discussed in the previous ap-
[cf. KLMTV Eq. (104)]. The contribution to the noise is pendix. However, the higher-order modes do not simply
then get dissipated — they too will propagate inside the in-
 
√ ñOPC1 terferometer (although with a different propagation law).
NOPC = εOPC , (B19)
ñOPC2 As a consequence, the exchange of energy between fun-
damental and higher modes due to mode-mismatching
where the ñOPCj are linear combinations of the individual
is coherent, and the formalism we have been using for
(independent) vacuum fields entering at each location (so
the loss does not apply. In this section, we shall extend
the spectral densities of these fields are unity and there
our formalism to include one higher-order mode and give
are no cross-correlations) and propagated to the output
an extremely simplified model of the mode-mismatching
port. KLMTV assumed that each of these losses is about
effects10 .
0.001, giving εOPC ∼ 0.003.
In a conventional interferometer (LIGO-I), the mode-
We next turn our attention to optical filters on the out-
mismatching comes predominantly from the mismatch of
put (as in the case of frequency-dependent homodyne de-
the mirror shapes between the two arms, which makes
tection for a squeezed-variational speed meter, discussed
the wavefronts from the two arms different at the beam
in Sec. IV B). Such cavities will have losses that may
splitter. In particular, the cancellation of the carrier
contribute significantly to the noises of QND interfer-
light at the dark port is no longer perfect, and addi-
ometers, as has been seen in KLMTV. In their Sec. VI,
tional (bright-port) noises are introduced into the dark-
KLMTV carried out a detailed analyses of such losses;
port output. For our speed meter, a third cavity—the
our investigation is essentially the same as theirs.
sloshing cavity—has to be matched to the two arm cavi-
The loss in the optical filters can come from scattering
ties, further complicating the problem.
or absorption in the cavity mirrors, which can be mod-
In order to simplify the situation, we approximate all
eled by attributing a finite transmissivity Te to the end
the waves propagating in the corner station (the region
mirrors, as we did for the arm cavities. The effect of lossy
near the beam splitter, where the distances are short
filters is again analogous to [Eq. (50)]. This time the loss
enough that ) as following the same phase-propagation
coefficient EF (ω) does have some frequency dependence:
law as a plane wave. The only possible source of mis-
X 1 X match is assumed to come from the difference of wave-
EF = 2εmm + ĒJ = 2εmm + (EJ+ + EJ− ) ,
2 front shapes (to first order in the fractional difference of
J=I,II J=I,II
the radii of curvature) and waist sizes for the light beams
(B20) emerging from the two arm cavities and the sloshing cav-
where εmm ∼ 0.001 is the mode-mismatching into each ity. Suppose, in the region of the corner station, we have
filter cavity and where a fiducial fundamental Gaussian mode Ψ(0) (which is be-
ing pumped by the carrier) with waist size w0 and wave-
4Te front curvature α0 ≡ 1/R0 that is roughly the same as
EJ± = (B21)
TJ [1 + (±ω/δJ − ξJ )2 ] those of the three cavities11 :
are the loss coefficents of the two different filter cavities   p
(0) 1 ρ2 α0 ρ2
(J = I, II) [cf. Eqs. (103) and (106) of KLMTV]. The Ψ (x, y) ∝ exp − 2 + ik , ρ = x2 + y 2 .
noise contribution is w0 w0 2
p  ñF1  (C1)
NF = EF . (B22)
ñF2
The weak frequency-dependence of EF will be neglected
(as KLMTV did), giving 10
This way of modeling the mode-mismatching effects was
εF ≃ EF ∼ 0.005 (B23) suggested to us by Stan Whitcomb.
11
We have chosen to use the curvature instead of the radius
[cf. Eqs. (107) and (104) of KLMTV]. The value of εF
of curvature because in this region the wavefronts are very
may vary slightly for the different optimizations we have
flat.
used, but it remains less than 0.006.

22
At leading order in the mismatches, the fundamental propagate them separately and then combine them. The
modes of the three cavities (in the region of the cor- input–output (a–b) relation of one of the cavities with
ner station), which have waist sizes wJ and curvatures mirrors held fixed can be written as
αJ ≡ 1/RJ [J =n, e, or slosh (for the north arm, east  (0)   
arm, and sloshing cavity, respectively)], can be written b̃  iΦ iΦexc
 ã(0)
= e fnd
Pfnd + e Pexc , (C6)
in the form: b̃(1) ã(1)

ΨJfnd (x, y) where

   
1 αJ − α0 ρ2 α0 ρ2  
∝ exp ikw02 exp − 2 + ik 1

w0 4 w0 2 Pfnd = 1 µ∗ , (C7a)
( µ
   
wJ − w0 αJ − α0 −µ∗

× 1+ + ikw02 Pexc = −µ 1 , (C7b)
4w0 16 1
" √ ! √ !#)
2x 2y are the projection operators, and Φfnd and Φexc = π are
× H2 + H2 , (C2)
w0 w0 the phases gained by the fundamental mode and excited
mode after being reflected back by the cavity.
where H2 (u) is the second-order Hermite polynomial of The mode-mismatching can cause both shot and radi-
u. This ΨJfnd (π, y) can be expressed as Ψ(0) plus a small ation pressure noises at the output, giving:
admixture of a higher-order mode Ψ(1) , which consists of
rad pres
equal amounts of TEM02 and TEM20 modes [and thus q̃(0) → q̃(0) + Nshot
MM + NMM . (C8)
is orthogonal to Ψ(0) ]. This admixture changes the waist
size from ω0 to ωJ and the curvature from α0 to αJ . We Assuming the mirrors are held fixed and applying the new
can choose our fiducial fundamental mode Ψ(0) in such a input–output relations (C6) of the non-perfect cavities,
way that the two arm cavities have an opposite mismatch we get the following shot noise in the output (to leading
with it, i.e., αn +αe = 2α0 , wn +we = 2w0 , and at leading order in µarm and µslosh ):
order, r p
 n, e     (0)  shot iψ ∗ 4 Tp
Ψfnd 1 ±µarm Ψ NMM = −e µarm p
= , (C3) To 1 + 1 − Tp
n, e
Ψexc ∓µ∗arm 1 Ψ(1) √
1 − 1 − Ti ωδ (1)
where “exc” denotes the excited mode and the admixing × √ ĩ
Ti |L(ω)|
amplitude µarm is, in general, complex. We also denote r
the fundamental and excited modes of the sloshing cavity −iψ ∗ Ti Tp ωδ (1)
≈e µarm ĩ ; (C9)
as 4To |L(ω)|
 slosh     (0) 
Ψfnd 1 µslosh Ψ see Eq. (B1). The quantity ĩ(1) refers to the excited mode
= ; (C4)
Ψslosh
exc −µ∗slosh 1 Ψ(1) of the noise coming in the bright port [I(ζ) in Fig. 3].
again, µslosh can be complex. We shall also assume that The main results embedded in Eq. (C9) are
the higher-order modes involved here are far from reso- (i) the mode-mismatching with the sloshing cavity
nance inside the cavities and will be rejected by them, does not give any contribution at leading order in
gaining a phase of π upon reflection from each cavity’s µ, and
input mirror. In the output, we assume the mode Ψ(0)
is selected for detection. (The local oscillator associated (ii) the mode-mismatching shot noise comes from the
with the homodyne detection is chosen to have the same higher-order mode entering from the bright port,
spatial mode as Ψ(0) , thereby “selecting” Ψ(0) . Note that strongly suppressed by the presence of the internal
the potential mode-mismatch effect here is already taken and power-recycling mirrors.
into account in the fractional loss εlo of the local oscilla-
tor, as described in Sec. B 4.) These two effects are both due to the coherent interac-
Quite naturally, we have to introduce two sets of tion between the fundamental (Ψ(0) ) and excited (Ψ(1) )
quadrature operators to describe the two modes. For ex- modes (of our idealized cavity), in which energy is not
ample, for the field P (ζ) entering through the extraction simply dissipated from Ψ(0) but exchanged coherently
mirror, we have between the two modes as the light flows back and forth
! ! between the sloshing cavity and the arm cavities. Detect-
(0) (1)
(0) p̃1 (1) p̃1 ing an appropriate linear combination of the two modes
p̃ ≡ (0) , p̃ ≡ (1) . (C5) can then be expected to reverse the effect of mode mis-
p̃2 p̃2
matching. In our case, the properties of the cavities are
For each of the three cavities, we have to decompose the carefully chosen such that Ψ(0) itself is the desired de-
optical field into its own fundamental and excited modes, tection mode (for the sloshing mismatch). Consequently,

23
the mode mismatching with the sloshing cavity does not gravity-wave sideband cancel. If, however, the transmis-
contribute at leading order [item (i) above]. Regarding sivity of the internal mirror, Ti , is not perfectly matched
item (ii), the mismatch of the two arm cavities does give by that of the RSE mirror, TRSE , then this cancellation
rise to an additional noise, but it can only come from the will no longer be perfect. As a result, the RSE cavity
higher mode in the bright port, because at leading or- (i.e., the cavity between the internal and RSE mirrors)
der in mismatches, (a) the propagation of Ψ(0) from the will have the same effect as an additional mirror (with
bright port to the dark port is suppressed and (b) there a small reflectivity). Suppose the transmissivity of this
is no propagation of dark-port Ψ(1) into dark-port Ψ(0) effective mirror is TRSE = (1 + εRSE )Ti . Then a simple
since we have chosen Ψ(0) in such a way that the two arm calculation yields its (amplitude) reflectivity:
cavities have exactly opposite mismatches with it. √ √
The reason why this noise is suppressed by the factor 1 − Ti − 1 − TRSE εRSE εRSE
µ= √ √ ≈ √ ≈ . (D1)
1/Tp is simple: because Ψ(1) is not on resonance with the 1 − 1 − Ti 1 − TRSE 2 1 − Ti 2
composite cavity formed by the power-recycling mirror
Adding this effective mirror with reflectivity µ to our
and the arm cavities, its fluctuations inside the system
interferometer yields a new set of input–output relations
(like its classical
p component) are naturally suppressed by
similar to Eq. (12), but with modified κ and ψ. The
a factor 1/ Tp compared to the level outside the cavity.
functional form of κ can be maintained by appropriately
The reason for the factor of 1/Ti is similar: the Ψ(1) redefining the quantities Ω and δ. To leading order in µ,
mode does not resonate within the system formed by the we obtain
arm cavities and the RSE mirror and will consequently
be suppressed. Ω3I δTM
κ → κTM = , (D2)
By computing at the fields at the end mirrors and (ω 2 − Ω2TM )2 + ω 2 δTM
2

from them the fluctating radiation pressure, we obtain

the radiation-pressure noise due to mode-mismatching: with
r   Ω → ΩTM = (1 − µ)Ω , δ → δTM = (1 − 2µ)δ . (D3)
rad pres e2iψ ∗ Ti Tp 0 0
NMM =− µarm ĩ(1) . (C10)
2 4To −κ∗ 0 Consequently, we can re-optimize the system to compen-
This radiation-pressure noise is suppressed by a factor sate for this transmissivity-mismatch effect.
similar to the shot noise.
By comparing Eqs. (C9) and (C10) with, e.g.,
Eqs. (B7), we see that mode mismatching produces noise
with essentially the same form as optical-element losses
from the arms, extraction mirror and sloshing cavity
(AES), with (assuming the input laser is shot-noise lim- [1] C. M. Caves, Phys. Rev. Lett. 45, 75 (1980).
ited in the higher modes) [2] A. Abramovici et al., Science 256, 325 (1992).
Ti Tp ∗ 2 [3] B. C. Barish and R. Weiss, Physics Today 52, 44 (1999).
εMM = |µarm | . (C11) [4] W. G. Unruh, in Quantum Optics, Experimental Grav-
4
itation, and Measurement Theory, eds. P. Meystre and
The factor Ti TP /4 happens to be the ratio between the in- M. O. Scully (Plenum, 1982), p. 647.
put power (at the power-recycling mirror) and the circu- [5] M. T. Jaekel and S. Reynaud, Europhys. Lett. 13, 301
lating power, which will be ∼ 10−4 . Suppose ℜ(µarm ) ∼ (1990).
ℑ(µarm ) ∼ 0.03. The effect of mode-mismatching will [6] E. Gustafson, D. Shoemaker, K. Strain, and R.
then be much less significant (in our simple model) than Weiss, LSC White Paper on Detector Research and
the losses from the optical elements. Development, LIGO document T990080-00-D (1999);
It should be evident that other imperfections in the available along with other relevant information at
cavity mirrors, which cause admixtures of other higher- http://www.ligo.caltech.edu/∼ligo2/ .
order (“excited”) modes, will lead to similar “dissipation [7] H. J. Kimble, Yu. Levin, A. B. Matsko, K. S. Thorne,
TT and S. P. Vyatchanin, Phys. Rev. D 65, 022002 (2001);
factors,” EMM ∼ i4 p |µ∗arm |2 . For this reason, we expect
cited in text as KLMTV.
mode mismatching to contribute negligibly to the noise, [8] P. Purdue, Phys. Rev. D 66, 022001 (2002); cited in text
and we ignore it in the body of the paper. as Paper I.
[9] V. B. Braginsky and F. Ya. Khalily, Phys. Lett. A 218,
167 (1996); V. B. Braginsky, M. L. Gorodetsky, and F.
APPENDIX D: TRANSMISSIVITY MISMATCH Ya. Khalili, Phys. Lett. A 232, 340 (1997).
BETWEEN THE INTERNAL MIRROR AND THE [10] V. B. Braginsky, M. L. Gorodetsky, and F. Ya. Khalili,
RSE MIRROR Phys. Lett. A 246, 485 (1998); quant-ph/9806081.
[11] F. Ya. Khalili, qr-qc/0203002.
Recall from Sec. I that when the internal and RSE [12] V. B. Braginsky and F. Ya. Khalili, Phys. Lett. A 147,
mirrors have the same transmissivity, their effects on the 251 (1990).

24
[13] V. B. Braginsky, M. L. Gorodetsky, F. Ya. Khalili, and
K. S. Thorne, Phys. Rev. D 61, 044002 (2000); cited in
text as BGKT.
[14] C. M. Caves, Phys. Rev. D 23, 1693 (1981).
[15] J. Mizuno, Ph.D. thesis, Max Planck Institute für Quan-
tenoptik, 1995.
[16] A. Freise, G. Heinzel, K. A. Strain, J. Mizuno, K. D.
Skeldon, H. Lück, B. Wilke, R. Schilling, A. Rüdiger,
W. Wingler and K. Danzmann, Phys. Lett. A 277, 135
(2000)
[17] J. Mason, Ph.D. thesis, California Institute of Technol-
ogy, 2001; also available as LIGO document P010010-00-
R at www.ligo.caltech.edu/docs/P/P010010-00.pdf .
[18] C. M. Caves and B. L. Schumaker, Phys. Rev. A 31, 3068
(1985).
[19] B. L. Schumaker and C. M. Caves, Phys. Rev. A 31, 3093
(1985).
[20] V. B. Braginsky, M. L. Gorodetsky, F. Ya. Khalili and
K. S. Thorne, “Energetic Quantum Limit in Large-Scale
Interferometers,” in Gravitational Waves, Proceedings of
the Third Edoardo Amaldi Conference, AIP Conference
Proceedings Vol. 523, ed. Sydney Meshkov (American In-
stitute of Physics, 2000) pp. 180–192; gr-qc/9907057.
[21] M. Xiao, L.-A. Wu and H. J. Kimble, Phys. Rev. Lett.
59 278 (1987).
[22] P. Grangier, R. E. Slusher, B. Yurke and A. La Porta,
Phys. Rev. Lett. 59, 2153 (1987).
[23] K. S. Thorne (private communication).
[24] S. P. Vyatchanin and A. B. Matsko, JETP 77, 218
(1993).
[25] S. P. Vyatchanin and E. A. Zubova, Phys. Lett. A 203,
269 (1995).
[26] S. P. Vyatchanin and A. B. Matsko, JETP 82, 1007
(1996).
[27] S. P. Vyatchanin and A. B. Matsko, JETP 83, 690
(1996).
[28] S. P. Vyatchanin, Phys. Lett. A 239, 201 (1998).

25