KAGRA Broadband Design Overview

• KAGRA broadband design is a 3 km dual-recycled, cryogenic interferometer using sapphire test masses at ~20–23 K to minimize thermal noise.
• The design employs resonant sideband extraction and variable RSE modes to balance shot noise and radiation-pressure noise across 10 Hz to a few kHz.
• Integration of advanced readout techniques, precise output mode cleaning, and cryogenic constraints enables upgrade paths like frequency-dependent squeezing.

KAGRA broadband design denotes the set of optical, cryogenic, mechanical, and readout choices by which the 3 km underground KAGRA interferometer is kept sensitive from roughly a few tens of hertz to a few kilohertz while remaining optimized for compact-binary science. In KAGRA, sapphire test masses operated at about $20$–$23\,\mathrm{K}$ reduce mirror thermal noise so strongly that much of the observation band becomes quantum-noise limited; as a result, broadband design is governed chiefly by how shot noise and radiation-pressure noise are redistributed through resonant sideband extraction, readout quadrature, and planned squeezing-based upgrades (Somiya, 2019, Akutsu et al., 2020).

1. Interferometer architecture and the meaning of broadband

KAGRA is a dual-recycled Fabry–Perot–Michelson interferometer with 3 km arm cavities, located underground in the Kamioka mine and operated with cryogenic sapphire test masses. Its baseline optical topology comprises power recycling, signal recycling in resonant sideband extraction (RSE), an input mode cleaner, and an output mode cleaner. In the overview literature, this combination of second-generation interferometer topology with underground siting and cryogenic optics is the basis for the description of KAGRA as a “2.5-generation detector” (Akutsu et al., 2018, Akutsu et al., 2020).

Within KAGRA design studies, “broadband” does not mean a perfectly flat noise spectrum. It means that the detector is not optimized for a single narrow resonance, but instead maintains low strain noise over a wide band extending from the low-frequency region limited by seismic, Newtonian, and suspension thermal noise, through the mid band important for binary neutron star inspirals, to the shot-noise-dominated high-frequency region. Early configuration studies explicitly treated the interferometer as a variable-RSE system, with a broadband tuned-RSE mode and a mildly narrow-banded detuned-RSE mode selected by control offsets rather than hardware replacement (1111.7185, Aso et al., 2013).

Several design summaries state the same target in slightly different forms. The 2018 overview gives a target strain sensitivity around $100\,\mathrm{Hz}$ of $2\times10^{-24}\,\mathrm{Hz}^{-1/2}$, with broadband sensitivity from $\sim 10$ to $2000\,\mathrm{Hz}$ optimized for compact binary coalescences. The 2020 design overview emphasizes a full “binary neutron star” band of $\sim 10\,\mathrm{Hz}$ to a few kilohertz, with the bucket region centered in the mid band rather than at a single optical resonance (Akutsu et al., 2018, Akutsu et al., 2020).

The auxiliary optics are part of that broadband definition rather than peripheral subsystems. The input mode cleaner in the baseline design has a 53.3 m round-trip length and finesse 540, while the output mode cleaner is a bow-tie cavity of about 1.5 m round-trip length and finesse $\approx 800$. These cavities are used to suppress laser spatial and frequency noise on the input side and to reject RF sidebands and higher-order modes at the antisymmetric port before DC readout (Akutsu et al., 2020).

2. Quantum-noise dominance as the organizing principle

In KAGRA, cryogenic operation pushes coating and substrate thermal noise below quantum noise over much of the observation band. This makes broadband design primarily a problem of quantum-noise shaping rather than of thermal-noise avoidance. The relevant decomposition is the standard one: photon shot noise dominates at high frequencies, while quantum radiation-pressure noise dominates at low frequencies. Because KAGRA uses sapphire mirrors of about $22.8$–$23\,\mathrm{kg}$, radiation-pressure noise is comparatively important at low frequency (Somiya, 2019, Akutsu et al., 2020).

The standard quantum limit enters explicitly in KAGRA design papers. For the differential arm mode,

$23\,\mathrm{K}$0

and the optomechanical coupling is written as

$23\,\mathrm{K}$1

With conventional phase-quadrature readout, the quantum-noise spectrum has the familiar structure

$23\,\mathrm{K}$2

so increasing power suppresses shot noise and simultaneously enhances radiation-pressure noise (Somiya, 2019, Michimura et al., 2020).

The role of signal recycling is therefore not secondary. In RSE, the effective arm power and bandwidth become

$23\,\mathrm{K}$3

so the interferometer can emulate a higher-power, broader-band response without imposing the same thermal load on the cryogenic mirrors. One study states explicitly that with low arm transmittance $23\,\mathrm{K}$4 and $23\,\mathrm{K}$5, the required beamsplitter power is about 25 times lower than for a simple Fabry–Perot Michelson while preserving the same SQL-touching behavior (Somiya, 2019).

This design logic is central to KAGRA. Since mirror temperature must remain near $23\,\mathrm{K}$6–$23\,\mathrm{K}$7, brute-force power scaling is limited by heat extraction through the cryogenic suspensions. Broadband performance is therefore obtained by shaping quantum noise through RSE, readout quadrature, and later squeezing, rather than by raising circulating power until shot noise alone is minimized (Somiya, 2019, Khalaidovski et al., 2014).

3. Variable RSE, back-action evasion, and optical-spring shaping

KAGRA was designed from the outset as a variable-RSE interferometer. In the broadband configuration, the signal-recycling cavity is tuned, giving broadband resonant sideband extraction (BRSE). In the detuned configuration, a small offset of the signal-recycling cavity creates an optical spring and redistributes sensitivity toward lower frequencies important for neutron-star inspirals (1111.7185, Aso et al., 2013).

In tuned RSE, the broadband effect is obtained by modified effective coupling rather than by a narrow resonance. In detuned RSE, the denominator of the full input–output relation develops optomechanical resonances, and the optical spring can produce a dip below the SQL near the spring frequency. The KAGRA design literature repeatedly describes the baseline detuning as small: $23\,\mathrm{K}$8. The point of that choice is not maximal narrowband gain, but improved inspiral range without excessive loss of bandwidth (Somiya, 2019, Ueda et al., 2014).

Back-action evasion (BAE) is the second built-in shaping mechanism. KAGRA detects a rotated output quadrature,

$23\,\mathrm{K}$9

so the readout angle $100\,\mathrm{Hz}$0 introduces a controlled correlation between shot noise and radiation-pressure noise. In tuned RSE, choosing $100\,\mathrm{Hz}$1 cancels part of the back-action over a limited band; in the design studies this was treated as a broadband quantum non-demolition technique rather than as a purely narrowband SQL-beating trick (Somiya, 2019, 1111.7185).

Reported binary-neutron-star ranges illustrate how strongly the broadband curve depends on configuration assumptions and optimization targets.

Study	Configuration	Reported BNS range
(1111.7185)	Tuned RSE, conventional readout	196 Mpc
(1111.7185)	Tuned RSE with BAE	206 Mpc
(1111.7185)	Detuned RSE, $100\,\mathrm{Hz}$2	238 Mpc
(Aso et al., 2013)	BRSE	217 Mpc
(Aso et al., 2013)	DRSE	237 Mpc
(Somiya, 2019)	Tuned RSE, phase readout	128 Mpc
(Somiya, 2019)	Tuned RSE, optimized $100\,\mathrm{Hz}$3	135 Mpc
(Somiya, 2019)	Detuned RSE, optimized $100\,\mathrm{Hz}$4	153 Mpc

These values come from different optimization studies and assumptions. What is consistent across them is the design principle: KAGRA does not use large detuning to maximize a very narrow dip. It uses moderate SRM reflectivity, moderate arm finesse, and small detuning so that both tuned and detuned modes remain scientifically useful. The 2011 and 2013 downselection papers therefore treat BRSE and DRSE as complementary operating modes rather than as mutually exclusive detector identities (1111.7185, Aso et al., 2013).

4. Readout chain, output mode cleaning, and technical-noise control

Broadband quantum-noise shaping in KAGRA is inseparable from the readout chain. The baseline readout is DC readout at the antisymmetric port, using a small carrier leakage as local oscillator. In simulations with realistic mirror maps, about 80% of combinations produced amplitude-quadrature leakage of at least $100\,\mathrm{Hz}$5 for $100\,\mathrm{Hz}$6 at the beamsplitter, and the 780 W configuration was stated to be comparable or larger. This is why KAGRA chose DC readout first rather than balanced homodyne detection, even though balanced homodyne would permit more flexible $100\,\mathrm{Hz}$7 control (Somiya, 2019).

That small local oscillator makes the output mode cleaner unusually important. The OMC requirements were set so that shot-noise-limited sensitivity would not degrade by more than 5% relative to an ideal OMC. For $100\,\mathrm{Hz}$8 of DC light before the OMC, the representative design allowances were TEM$100\,\mathrm{Hz}$9 loss $2\times10^{-24}\,\mathrm{Hz}^{-1/2}$0, residual higher-order mode power $2\times10^{-24}\,\mathrm{Hz}^{-1/2}$1, and residual RF sideband power $2\times10^{-24}\,\mathrm{Hz}^{-1/2}$2. The selected OMC parameters were a four-mirror bow-tie cavity with half-round-trip length $2\times10^{-24}\,\mathrm{Hz}^{-1/2}$3, round-trip Gouy phase $2\times10^{-24}\,\mathrm{Hz}^{-1/2}$4, and finesse below 800. In a pessimistic but acceptable mirror-map realization, the difference in shot-noise-limited strain sensitivity at $2\times10^{-24}\,\mathrm{Hz}^{-1/2}$5 between ideal and realistic OMC models was 3.2% (Kumeta et al., 2014).

Detuned operation introduces a separate technical-noise problem. In detuned RSE, the $2\times10^{-24}\,\mathrm{Hz}^{-1/2}$6 PM sidebands used for SRCL sensing are tilted from the PM axis and made unequal by the 50.5/49.5 beamsplitter asymmetry, which couples photo-detector noise (PDN) and oscillator phase noise (OPN) into the DARM channel. The proposed cure was to add an amplitude-modulated $2\times10^{-24}\,\mathrm{Hz}^{-1/2}$7 sideband with adjustable depth and phase so that the total field seen by the interferometer again behaves as a pure, balanced PM pair (Ueda et al., 2014).

Optickle scans in that study found a deep PDN minimum near relative AM amplitude $2\times10^{-24}\,\mathrm{Hz}^{-1/2}$8 and relative phase $2\times10^{-24}\,\mathrm{Hz}^{-1/2}$9, and an OPN minimum near relative AM amplitude $\sim 10$0 and relative phase $\sim 10$1. The operating point emphasized OPN suppression: around 65% AM amplitude and $\sim 10$2 relative phase. At $\sim 10$3, this reduced OPN by a factor of about 192 relative to uncompensated detuned RSE, while keeping PDN below the design curve. The additional modulator loss could be kept to about 5% with $\sim 10$4, $\sim 10$5, and $\sim 10$6 (Ueda et al., 2014).

The practical implication is that KAGRA broadband design includes not only the quantum-noise spectrum itself, but also the RF and output-optics infrastructure needed to keep that spectrum observable in detuned operation. The OMC, DC local-oscillator budget, and AM-sideband compensation are therefore part of the broadband design rather than post hoc technical corrections (Kumeta et al., 2014, Ueda et al., 2014).

5. Cryogenic and underground constraints as broadband determinants

KAGRA’s cryogenic and underground features reduce classical noise, but they also impose the boundary conditions that define the optical design. The underground site lowers seismic and Newtonian noise; design papers cite seismic motion in the Kamioka mine as roughly 100 times smaller than at the TAMA site in the 1–100 Hz range, and later overviews describe the core suspensions as the world’s tallest vibration-isolation systems at 13.5 m. At the same time, the cryogenic payload fixes the admissible mirror temperature and therefore the allowable laser heat load (1111.7185, Akutsu et al., 2018).

The final suspension stage is especially constraining. KAGRA uses four sapphire fibers connecting a $\sim 10$7 test mass to an intermediate mass at about $\sim 10$8. In the heat-extraction study, the baseline geometry was length $\sim 10$9, diameter $2000\,\mathrm{Hz}$0, with the conductivity requirement

$2000\,\mathrm{Hz}$1

Measured monolithic and annealed HEM sapphire fibers achieved $2000\,\mathrm{Hz}$2 of $2000\,\mathrm{Hz}$3–$2000\,\mathrm{Hz}$4 and about $2000\,\mathrm{Hz}$5, respectively, showing that the basic heat-conduction requirement was technologically feasible (Khalaidovski et al., 2014).

The absorbed-power budget is what couples this directly to broadband sensitivity. Under optimistic assumptions, the total absorbed power in an ITM was estimated at about $2000\,\mathrm{Hz}$6; under more conservative assumptions it could reach about $2000\,\mathrm{Hz}$7. Prototype tests scaled to full length implied roughly $2000\,\mathrm{Hz}$8 extraction through four fibers in a realistic welded-fiber case and about $2000\,\mathrm{Hz}$9 in an idealized case. That is sufficient for the optimistic heat load but not for the conservative one, in which case the input laser power would have to be reduced to about 60% of maximum to keep the ITM below $\sim 10\,\mathrm{Hz}$0 (Khalaidovski et al., 2014).

This thermal constraint feeds directly into the broadband noise curve. Thicker fibers improve heat extraction, since heat-extraction capability scales as $\sim 10\,\mathrm{Hz}$1, but they also increase suspension thermal noise, which scales as $\sim 10\,\mathrm{Hz}$2. The cryogenic design therefore fixes a trade-off: raising power suppresses shot noise, but the resulting heat load forces thicker or shorter fibers, which worsen suspension thermal noise and can raise mirror temperature. KAGRA’s broadband design is the chosen compromise among those coupled constraints, not an optical design imposed independently of cryogenic engineering (Michimura et al., 2020, Khalaidovski et al., 2014).

The operating temperature itself was also optimized as a broadband parameter. The 2020 overview shows sensitivity curves for mirror temperatures of 22, 50, 100, 150, and 300 K and identifies 22 K as the nominal choice because it minimizes the combined thermal budget while remaining feasible for cooling. In that sense, the cryogenic mirror temperature is as much a broadband design parameter as SRM reflectivity or arm finesse (Akutsu et al., 2020).

6. Upgrade paths and evolving meanings of broadband

Planned and proposed upgrades extend the same design logic. The 2019 quantum-noise review identifies frequency-independent squeezing as a low-risk step and frequency-dependent squeezing (FDS) through a filter cavity as the key KAGRA+ broadband technology. In the idealized discussion, 10 dB FDS with no losses would act like an approximately tenfold increase in effective intra-arm power and an effective increase in test mass by about $\sim 10\,\mathrm{Hz}$3, thereby reducing both shot noise and radiation-pressure noise without exceeding the cryogenic heat budget. The associated filter-cavity condition was expressed by matching the cavity pole to the radiation-pressure SQL-touching frequency, and a representative example stated that if $\sim 10\,\mathrm{Hz}$4, a 30 m filter cavity would require finesse $\sim 10\,\mathrm{Hz}$5 (Somiya, 2019).

The 2020 KAGRA upgrade study made the broadband concept explicit in parameter form. Its “combined” design used 100 kg sapphire mirrors, $\sim 10\,\mathrm{Hz}$6 input power at the beamsplitter, nearly tuned SRC with detuning $\sim 10\,\mathrm{Hz}$7, SRM reflectivity $\sim 10\,\mathrm{Hz}$8, homodyne angle $\sim 10\,\mathrm{Hz}$9, and 10 dB squeezing through a 30 m filter cavity. The reported gains were a BNS range of 302 Mpc versus 153 Mpc for the default design, a $\approx 800$0–$\approx 800$1 BBH range of 707 Mpc versus 353 Mpc, and median sky localization of $\approx 800$2 versus $\approx 800$3 for a GW170817-like event in an HLVK network. The same study contrasted that broadband design with low-frequency, high-frequency, larger-mirror, and FD-squeezing-only configurations, each optimized for a narrower science goal (Michimura et al., 2020).

The science white paper framed those upgrade options by source class. It divided the band into low frequencies of about $\approx 800$4–$\approx 800$5, middle frequencies of about $\approx 800$6–$\approx 800$7, and high frequencies of about $\approx 800$8 to $\approx 800$9, and concluded that no single option improves all science cases because the frequency band of the sensitivity improvement is not broad enough. The “40 kg” and “FDSQZ” options were therefore treated as the more nearly broadband upgrade paths, while LF and HF options were treated as frequency-tilted designs (collaboration et al., 2020).

A further reinterpretation appears in the 2025 decadal strategy. That study compared 14 post-O5 upgrade options and concluded that the technically preferred direction is not a uniformly improved full-band detector, but a high-frequency moderate upgrade, HFmod, that enhances sensitivity over a broad band above about $22.8$0. In the reported network science metrics, this strategy would enable sky localization of BNS mergers at $22.8$1 to better than $22.8$2 in a LIGO–Virgo–KAGRA network and improve the measurement precision of the tidal deformability parameter by approximately 10% at median compared to a network without KAGRA (collaboration et al., 5 Aug 2025).

Taken together, these studies define KAGRA broadband design as a moving but coherent concept. In the baseline detector it means a cryogenic, underground, variable-RSE interferometer whose parameters are chosen so that quantum noise is shaped broadly enough for compact-binary astronomy under stringent thermal constraints. In KAGRA+ and post-O5 planning it means either full-band quantum-noise reduction through heavier mirrors, higher power, and frequency-dependent squeezing, or, in the more recent strategic sense, a deliberately high-frequency-biased broadband response above $22.8$3 that maximizes KAGRA’s contribution to network timing, localization, tidal measurements, and continuous-wave searches (Somiya, 2019, collaboration et al., 5 Aug 2025).