LIGO A: Detector Design and Science Prospects Beyond A+

The paper presents the LIGO A detector concept, an upgrade for the LIGO observatories based on room-temperature interferometers beyond the fifth observing run (O5). Building on the A+ sensitivity, A targets broadband sensitivity improvements through heavier test masses, improved suspensions and seismic isolation, increased arm-cavity power, enhanced frequency-dependent squeezing, reduced coating thermal noise considering two scenarios, and improved control of mechanical motion and optical modes. The paper describes the principal design choices, projected noise performance, and corresponding astrophysical prospects. LIGO A substantially increases compact-binary detection rates, strengthens population inference, and improves both early-warning times and localization for binary neutron star mergers. The improved sensitivity enables more detailed studies of compact-binary coalescences, including higher-order multipoles, intermediate-mass black holes, remnant black hole ringdown, and the neutron star equation of state. It also broadens the discovery potential for new gravitational-wave sources such as continuous waves and bursts, should enable detection of the stochastic background from compact binary mergers if it remains undetected after O5, and strengthens the role of gravitational-wave detectors as probes of fundamental physics. The paper discusses key technical challenges and the role of A as both a major scientific upgrade for the 2030s and a technology pathfinder for next-generation gravitational-wave observatories, such as Cosmic Explorer.

The A upgrades to the LIGO A+ detectors retain room-temperature operation, fused-silica test masses, and 1064 nm laser light. This leverages existing expertise and provides a natural path toward implementation in Cosmic Explorer detectors. The design changes can be divided into improvements targeting low frequencies (below 50 Hz), mid frequencies (50 to 300 Hz), and high frequencies (above 300 Hz).

The principal designs are as follows:

• A+ design: The A+ configuration builds on Advanced LIGO and corresponds to the projected sensitivity of the final stage of O5 (O5c). It incorporates a 30% reduction in coating thermal noise relative to the current Advanced LIGO level and implements 7 dB of observed frequency-dependent squeezing. The test masses are 40 kg, and the circulating laser power in each arm cavity is 550 kW.

• A baseline: Relative to A+, A baseline increases the test masses to 105 kg, redesigns the suspensions for the heavier optics and improved controllability, and raises the suspension-fiber stress by a factor of two to 1.6 GPa. The low-frequency design also includes upgraded seismic isolation sensors and a modest factor-of-two suppression of Rayleigh-wave Newtonian noise. At high frequencies, A baseline increases the arm-cavity circulating power to 1.5 MW and pushes the observed squeezing level to 10 dB. The coating thermal noise is reduced to 50% of the Advanced LIGO level.

• A wideband: Identical to A baseline, but with the signal bandwidth broadened to 3.4 kHz (from 450 Hz) by increasing the reflectivity of the signal recycling mirror, together with corresponding adjustments to the finesse and detuning of the squeezer filter cavity.

• A optimistic: The A baseline configuration defined above is limited by coating thermal noise at mid-frequencies, so lower-thermal-noise coatings would improve the interferometer performance and increase the science reach. As an optimistic limiting case, we consider a strain curve in which the coating thermal noise is equal to the A quantum noise at 100 Hz; this corresponds to a coating thermal noise reduction of approximately 75% relative to the Advanced LIGO level. The possibility of realizing this target with GaAs/AlGaAs crystalline coatings is discussed.

The low-frequency sensitivity improvements for A stem from the following changes to the interferometers: the test masses are increased in size and mass; the test-mass suspensions are redesigned to reduce thermal noise and control noise; the active seismic isolation platforms are upgraded with lower-noise sensors to reduce control noise; and ground-motion sensors are added to enable subtraction of Newtonian noise. These upgrades target performance improvements from 10 to 50 Hz, and maintain the performance increase even with very high-power operation.

The mass of the fused-silica test mass is increased to 105 kg (from 40 kg), with a diameter of 45 cm (from 34 cm) and a thickness of 30 cm (from 20 cm). The increased mass, m, has several benefits to the strain noise: quantum radiation pressure noise is reduced (as 1/m); suspension thermal noise is reduced (as 1/√m); the impact of gas damping force noise is reduced (as m−2/3 under certain assumptions); and the increased moment of inertia (by a factor of 4.6) counters radiation-pressure torques, helping to stabilize arm-cavity alignment.

The suspension fibers for A will operate at a stress of at least 1.6 GPa, twice that of the current LIGO fibers, with the potential to increase to 2 GPa. Suspension thermal noise is only marginally the dominant strain noise contributor from 10 Hz to 14 Hz with this design. The test-mass bounce and roll modes, at 6.6 Hz and 9.3 Hz respectively, are both below 10 Hz, and the fundamental violin-mode frequency is increased from 500 Hz to 740 Hz, reducing its impact in the mid-frequency detection band.

The four-stage suspension system has been redesigned to support the larger and heavier A test-mass mirrors. The total mass of the main suspended chain is increased to 420 kg (from 120 kg). Each of the four stages has approximately the same mass, in contrast to the current suspension, where the top two stages have half the mass of the lower two. This updated mass distribution keeps the suspension modes more closely spaced in frequency, which improves the passive isolation and reduces the required bandwidth of the local damping loops (thus reducing the infiltration of damping noise into the detector’s strain measurement). All the wires in the A suspension are vertical, whereas many Advanced LIGO wires are mounted at angles, to give more flexibility in setting the mode frequencies at the expense of increased cross coupling. The A suspension has four independent blades, each with a separate wire, supporting the three upper stages.

In the A suspension, improved sensors will be used throughout, with interferometric sensors on the main chain and DC sensitive beam deflection sensors (QOSEM) or BOSEM sensors on the reaction chain. The QOSEM achieves a noise floor of 4 pm/√Hz at 1 Hz, while using the same mounting and actuation scheme as the current shadow sensors. For the A suspension, the local sensor noise floor must be 5 pm/√Hz or less at 10 Hz, which all of these displacement sensors are able to achieve.

The A upgrade will continue to use the existing Advanced LIGO internal seismic isolation (ISI) system, with additional advanced sensors to improve the isolation performance in important frequency bands. One type of upgrade is to equip the ISI systems with new or additional, lower-noise inertial sensors. The other type of upgrade for the seismic system is the Seismic Platform Interferometer (SPI). The goal of the SPI is to stabilize the relative motion between seismic platforms.

Newtonian gravitational noise will limit the low-frequency sensitivity of the A detector when all improvements discussed above are implemented. The dominant source of Newtonian noise for the A detectors is Rayleigh seismic waves near the test masses. The target is a relatively modest suppression of Newtonian noise from Rayleigh waves by a factor of two. This can be achieved by combining an array of seismometers within the corner station with additional ground-tilt sensors underneath the individual test masses.

Improving the sensitivity significantly at mid-frequencies will require new test-mass coatings with lower thermal noise. For the A baseline coatings, we assume that the 50% reduction for titania-germania coatings suggested in Ref. [71] can be realized with further development and applied to both input and end test masses. Thus, the A baseline design incorporates this 50% reduction in coating thermal noise. To evaluate the science impact of better coatings, we consider a reduction in coating thermal noise by approximately a factor of two relative to the baseline case. At this level, coating thermal noise is roughly comparable to substrate thermal noise and quantum noise around 100 Hz. One coating technology that could potentially reach this lower-noise target is crystalline coatings made from GaAs/AlGaAs Bragg stacks.

Further increases in high-frequency sensitivity can be achieved through a combination of higher operating optical power and an increase in effective squeezed vacuum enhancement level. The operating optical-power goal for A is to have 1.5 MW resonantly stored in each arm cavity, about four times the level achieved during O4 and about three times the A+ design value. The goal for squeezed light enhancement is set at 10 dB of broadband, effective squeezing, matching the target specified for the Cosmic Explorer detector design and improving on the 6 dB realized during O4 and the 7 dB assumed for A+. To achieve 1.5 MW of circulating power in the arm cavities, the input laser would need to operate at 300 to 350 W. Solutions for reaching the required A laser power include fiber-based amplification and coherent combination of separately amplified lasers.

Achieving the A target of 10 dB frequency-dependent broadband squeezing enhancement will require several improvements over current squeezing implementations. Current squeezed-light sources used in GW detectors generate frequency-independent squeezed states from optical parametric oscillators (OPOs), which are then reflected from 300 m optical filter cavities, producing a frequency-dependent rotation of the squeezing angle. The dominant optical losses are currently from the filter cavity, readout chain, and the injection path. The effective losses in the OPO currently total to 2% in LIGO O4 run, and only incremental improvements are needed to reach the A goal of 1%. The installation of new low-loss output mode cleaners (OMCs) and minor improvements to the output steering optics are expected to reduce the readout loss from 4.5% to 3.5%. Suspended adaptive mode-matching stages (SAMS) are deformable mirrors used as telescopes to spatially match the OPO, OMC, filter cavity, and interferometer to each other.

The high-frequency sensitivity can also be improved by increasing the signal extraction mirror reflectivity in order to broaden the bandwidth of the interferometer. This A wideband configuration is shown by the purple trace in figure 1. Relative to the A baseline configuration, there are several trade-offs associated with using the wideband configuration. First, the mid-frequency sensitivity is significantly reduced, leading to a decrease in astrophysical signal detection range. Second, the filter-cavity finesse must be increased in order to match the lower-frequency crossover of the quantum noise, which makes the detector more susceptible to filter-cavity noise. Third, the detector is more susceptible to mode mismatch and optical loss in the SEC, since this loss is not broadened along with the other quantum-noise contributions.

Overall, A substantially extends the compact-binary reach relative to A+. For the fiducial systems considered here, the A baseline configuration increases the BNS range from 280 to 440 Mpc and the BBH range from 2.2 to 3.0 Gpc. It also increases the maximum horizon redshift for equal-mass binaries from 2.6 to 4.7, with further gains possible for the optimistic configuration with lower coating noise. Relative to A+, the A baseline configuration increases the expected annual detections by factors of several across all compact-binary classes, yielding catalogs with tens to hundreds of BNS and NSBH events and several thousands of BBH events per year. The A baseline configuration increases the expected annual detections by factors of several across all compact-binary classes, yielding catalogs with tens to hundreds of BNS and NSBH events and several thousands of BBH events per year.

Early warning of BNS mergers is one of the most direct science drivers for improved low-frequency sensitivity. The warning time is only about 30 s for the representative O4 sensitivity and 1.6 min for A+, but increases to approximately 6–7 min for the considered A configurations. The similar performance across the three A configurations indicates that this observable is driven primarily by the low-frequency design rather than by the mid- or high-frequency improvements.

For the A+ network, we find that O(14) events per year are localized to better than 100 deg2. The higher detection rates of the A networks increase this to O(26–43) events per year. Localization to better than 100 deg2, combined with the improved early-warning capability of A, would enable more rapid electromagnetic follow-up and strengthen the prospects for multi-messenger observations.

With the A network, it becomes possible to identify the peak of the BBH merger-rate evolution and constrain both its redshift and the overall shape of the redshift distribution. The A baseline network is able to constrain the position of the redshift peak well, while also providing constraints on the other hyperparameters that determine the shape of the distribution. These results demonstrate the potential of A for population inference at redshifts where current GW detector networks have limited constraining power.

The A baseline configuration improves notably on A+ for higher-order mode detection, while the A optimistic provides an even greater improvement. For a given SNR, A optimistic yields at least a factor-of-a-few increase in the fraction of events with detectable higher-order modes relative to the A+ or A configurations without major improvements to the coating thermal noise. The baseline A configuration would increase the annual number of higher-order-mode detections by a factor of approximately 1.5–1.8 relative to A+, while the A optimistic configuration could yield up to three times as many detections.

The key advantage of A for heavy systems such as GW231123 is its improved sensitivity below 20 Hz, which brings more of the inspiral of high-mass BBH mergers into band. If such a system was observed by an A detector network, the parameter uncertainties would be approximately one third of the current measurement, enabling tighter constraints on the binary’s astrophysical origin. The low-frequency improvements enable an informative measurement of χp for a GW231123-like BBH at a redshift of z = 1.6, while the measurement from 20 Hz recovers the prior. The observation of significant spin precession in an intermediate mass BBH at a redshift of z = 1.6 would provide compelling evidence for hierarchical or dynamically assembly operating efficiently at early cosmic times.

The improved sensitivity of A yields higher ringdown SNRs, making it possible to measure the dominant QNM more accurately, improve the prospects for resolving subdominant higher-order modes and overtones, and extend ringdown studies to a broader range of BBH systems. For a relatively low-mass BBH system with a total binary mass of 20 M⊙ at a luminosity distance of 400 Mpc, A baseline yields a ringdown SNR of 12.3, compared with 3.8 for O4 and 6.0 for A+. Among the detector configurations considered, only A yields informative mass and spin posteriors; by contrast, the O4 and A+ configurations produce broad, largely prior-dominated posteriors.

For neutron star science, the constraints on the binary tidal deformability Λ̃, as characterized by the 90% credible interval, improve by approximately 40% between A+ and various A configurations. The uncertainty on the radius of the primary neutron star is approximately 32% for A, constituting an improvement of around 8% over A+. For post-merger signals, the loudest 10% of events reach ρpm ∼ 3–4 for the A configurations, while the largest value among all simulations is ρpm < 6 and is obtained for the A wideband configuration. This represents a clear improvement over A+, for which the corresponding post-merger SNRs are of order unity, but the absolute SNRs remain too small for robust post-merger detection except possibly for the most favourable nearby events.

For continuous waves, the results indicate a substantial improvement in detection prospects relative to the current constraints, particularly for the A optimistic. The A configurations improve the sensitivity to the ellipticity by approximately a factor of two relative to A+ in a wide frequency range.

Any of the proposed upgrades to the A+ design would provide sufficient sensitivity to ensure detection of the stochastic background from binary mergers, should it remain undetected at the end of O5. Such a detection would provide an independent probe of the cosmic merger history and populations of compact binaries, while improved characterization of this astrophysical foreground would benefit searches for weaker stochastic backgrounds of cosmological or other astrophysical origin.

For beyond-Standard-Model particles and dark matter, the improved sensitivity of A relative to A+ extends the reach of searches for boson clouds around black holes. For a system like GW250114 with remnant BH mass 63 M⊙ and dimensionless spin 0.7, the horizon distance for a vector-boson superradiance search increases from 0.9 Gpc with A+ to 1.3 Gpc with A, corresponding to an enhancement of approximately 45% in distance and a factor of ∼3 in accessible volume. The gain is even larger for more optimal systems: for a remnant BH with mass ∼300 M⊙ and spin 0.9, the horizon distance increases from approximately 5.0 Gpc with A+ to 8.2 Gpc with A, corresponding to a factor of ∼4 increase in accessible volume. The broadband sensitivity improvement of A relative to A+ extends the reach of LIGO to a wider region of ultralight-particle parameter space and strengthens the role of ground-based GW detectors as probes of dark-sector physics.

The A concept provides a path for substantially improving the scientific reach of the LIGO observatories beyond O5 while remaining within the existing 4-km facilities. The design is deliberately broadband: rather than being tuned for a single source class or frequency band, it combines low-frequency improvements from heavier test masses, upgraded suspensions, and improved seismic isolation, with mid- and high-frequency gains from lower coating thermal noise, higher arm power, and improved frequency-dependent squeezing performance. This balanced approach enables a broad science program, including increased compact-binary detection rates, improved population inference, stronger access to the properties of astrophysical sources, and enhanced sensitivity to stochastic backgrounds, bursts, continuous waves, and new physics.

Realizing this performance will require addressing several key technical challenges. At low frequencies, the target sensitivity depends not only on reducing fundamental suspension and Newtonian noise, but also on suppressing technical noise associated with seismic motion, suspension damping, and global angular control. At mid-frequencies, improved coating thermal noise remains a major challenge, making continued progress in low-loss optical coatings essential. At high frequencies, achieving stable operation at higher circulating optical power and with improved squeezing performance will require further reductions in optical loss, improved mode matching, better control of thermal distortions, and robust suppression of angular and parametric instabilities.

Beyond its immediate scientific return, A serves an important strategic role in the global GW roadmap. Many of its core technologies, including heavier test masses, improved suspensions, low-noise coatings, higher optical power, improved squeezed-light injection, and advanced seismic and Newtonian-noise mitigation, are directly relevant to next-generation observatories such as Cosmic Explorer and the Einstein Telescope. Implementing these technologies in the existing LIGO facilities would therefore provide a critical opportunity to mature hardware, control schemes, commissioning strategies, and data-analysis methods before they are required at larger scale in future detectors, while also sustaining the specialized expertise essential for commissioning and operating next-generation observatories. In this sense, A is both a major scientific upgrade for the 2030s and a technology pathfinder toward next-generation GW observatories.

However, the A design remains constrained by the existing 4-km LIGO infrastructure and therefore cannot deliver the same cosmological volume or broad transformational science capability expected from new observatories such as Cosmic Explorer and the Einstein Telescope. Rather, A occupies a natural place between A+ and the next generation of GW detectors: it would provide an important intermediate step by delivering substantial science gains within the existing infrastructure before new facilities come online, while maturing technologies, commissioning experience, and data-analysis methods needed for the observatories that follow. By combining near-term feasibility with broad scientific reach, A provides a compelling route for advancing GW astronomy in the post-A+ era.

Improvements to AI Systems:

1. Gravitational-wave signal detection and classification AI: Train deep learning models on simulated A noise curves (baseline, wideband, optimistic) to detect and classify compact binary coalescences (BNS, BBH, NSBH) with higher sensitivity at low frequencies (10–50 Hz). The AI can learn to separate signals from Newtonian noise and suspension thermal noise, improving detection confidence for high-mass systems and intermediate-mass black holes.

2. Early-warning prediction AI: Develop recurrent or transformer-based models that use real-time low-frequency strain data (10–30 Hz) to predict binary neutron star merger times 6–7 minutes in advance (vs. 30 s currently). The AI can be trained on A-specific sensitivity improvements to issue alerts with quantified localization uncertainties (<100 deg2) for electromagnetic follow-up.

3. Parameter estimation AI: Build neural posterior estimators (e.g., normalizing flows) trained on A-simulated signals to infer binary parameters (masses, spins, tidal deformability) with reduced uncertainties—e.g., 40% tighter constraints on Λ̃ and 8% better neutron star radius estimates. The AI can handle higher-order multipoles and precessing spins at redshifts up to z=4.7, which current models struggle with.

4. Population inference AI: Use hierarchical Bayesian neural networks to infer the redshift evolution of BBH merger rates from A catalogs (thousands of events/year). The AI can constrain the peak of the merger-rate redshift distribution and distinguish between hierarchical and dynamical assembly scenarios using spin-precession measurements at z=1.6.

5. Anomaly detection for new physics AI: Train unsupervised anomaly detectors on A noise spectra to flag continuous waves from boson clouds, stochastic backgrounds, and burst signals. The AI can exploit the 45%–60% increase in accessible volume for ultralight particle searches and identify deviations from standard-model predictions.

6. Control and noise mitigation AI: Implement reinforcement learning agents to optimize suspension damping, seismic isolation, and Newtonian-noise subtraction in real time. The AI can manage the 420 kg suspension chain, maintain alignment under radiation-pressure torques, and reduce technical noise from sensors (e.g., QOSEM at 4 pm/√Hz) to meet A sensitivity targets.

7. Optical system optimization AI: Use physics-informed neural networks to optimize squeezing injection (10 dB), filter-cavity detuning, and signal-recycling mirror reflectivity for wideband operation (3.4 kHz bandwidth). The AI can predict and correct for mode mismatch, thermal distortions, and parametric instabilities at 1.5 MW circulating power.

8. Multi-messenger coordination AI: Develop a decision-support AI that integrates A early-warning alerts with electromagnetic telescope scheduling, prioritizing follow-up for events with high localization confidence and high scientific value (e.g., BNS mergers with detectable post-merger signals, ρpm>3).

9. Technology pathfinder AI for next-gen detectors: Use A commissioning data to train transfer-learning models that predict performance of Cosmic Explorer components (e.g., 105 kg test masses, crystalline GaAs/AlGaAs coatings). The AI can simulate failure modes and optimize designs before physical implementation.

10. Real-time data quality AI: Create a classifier that distinguishes genuine astrophysical signals from non-stationary noise artifacts (e.g., suspension violin modes at 740 Hz, seismic disturbances). The AI can flag data segments for vetoing or glitch subtraction, improving the effective duty cycle of A observations.