Probing ultralight bosons with LISA observations of spinning black hole mergers and follow-up searches of merger remnants

This paper investigates the potential of the Laser Interferometer Space Antenna (LISA) to probe ultralight bosons through black hole superradiance, using two complementary observational strategies: measurements of black hole spins from merging massive black hole binaries, and targeted follow-up searches for gravitational wave signals from superradiant boson clouds around merger remnants.

The study considers three massive black hole population models: PopIII (a light-seed scenario with black holes forming at high redshift in the mass range 10 2–10 cubed M⊙), Q3delays (a heavy-seed scenario with black holes in the range 10 4–10 5 M⊙, including delays between galaxy and black hole mergers), and Q3nodelays (a similar heavy-seed scenario but without delays, predicting more events at higher redshift). The authors use the SuperRad package to model superradiant evolution and gravitational wave emission, and the lisabeta code with the IMRPhenomHM waveform model to compute signal-to-noise ratios (SNRs) for the binary merger signals. They adopt a detectability threshold of SNR ≥ 20 for binaries and SNR ≥ 10 for follow-up signals, assuming a four-year LISA mission.

For spin measurements, the authors compute the maximum spin a black hole can support given a boson mass and spin-down timescale, using SuperRad to evolve superradiant clouds for initial spins in [0,1). They consider spin-down timescales of the Salpeter time TS = 4.5 × 10 7 years and a more conservative 0.01 TS, and assume a relative spin uncertainty of 10% (also considering 1% for comparison). A boson mass is considered excluded if the measured spin exceeds the maximum allowed spin after accounting for measurement uncertainty.

The results show that spin measurements can constrain scalar boson masses in the range approximately [5 × 10−18, 10−14] eV and vector boson masses in the range approximately [6 × 10−19, 2 × 10−14] eV, with the exact ranges depending on the population model. Specifically, for a spin-down timescale of TS, the excluded mass ranges with probability > 0.99 are: for Q3delays, scalars [10−17, 2 × 10−15] eV and vectors [2 × 10−18, 3 × 10−15] eV; for Q3nodelays, scalars [5 × 10−18, 10−14] eV and vectors [6 × 10−19, 2 × 10−14] eV; for PopIII, scalars [3 × 10−17, 5 × 10−14] eV and vectors [3 × 10−18, 7 × 10−14] eV. The spin-down timescale of 0.01 TS narrows the exclusion region, particularly at the lower-mass end, because lower boson masses produce more extended clouds with longer instability growth times.

For follow-up searches, the analysis is restricted to vector bosons, as scalar cloud growth timescales typically exceed the LISA mission duration. The excluded mass ranges with probability > 0.99 are [4 × 10−17, 3 × 10−16] eV for Q3delays, [3 × 10−17, 2 × 10−15] eV for Q3nodelays, and [10−16, 3 × 10−15] eV for PopIII. These ranges are significantly narrower than those from spin measurements, because follow-up signals are quasi-monochromatic and generally weaker, so only a small fraction of remnants fall within the LISA sensitivity band with sufficient SNR.

The paper also considers the detection scenario where an ultralight vector boson exists and affects the binary components prior to merger through superradiant spin-down. In this case, the authors apply the spin-down procedure to all binary components satisfying the superradiance condition, recompute the merger remnant properties using fitting formulae, and then assess the detectability of follow-up signals. The detection probabilities are limited: for boson masses in the range 10−16–10−15 eV, the detection probability exceeds 20% for Q3delays and 40% for Q3nodelays, reaching ≳80% in the range [2 × 10−16, 1.5 × 10−15] eV for Q3nodelays. For PopIII, the detection probability exceeds 20% reaching 40% in the range [5 × 10−16, 1.5 × 10−15] eV. A key limitation is that the same superradiant instability responsible for generating post-merger signals also acts on the binary components before merger, reducing remnant spins and suppressing both the growth rate of the instability and the strength of the resulting gravitational wave signal.

The authors note that their scalar boson constraints from spin measurements are broadly consistent with earlier analyses by Brito et al. (2017), which reported expected constraints in the mass range [4 × 10−18, 10−14] eV and [10−18, 2 × 10−13] eV, though those studies used older catalogs. The paper also discusses limitations and future directions, including the potential impact of self-interactions for scalar fields (e.g., bosenova collapse) and couplings of vector bosons to other sectors (e.g., dark photon-photon kinetic mixing), estimating that quartic self-interaction couplings with f ≳ O(10 15–10 16) GeV would not affect the purely gravitational analysis, and that spin constraints would apply for kinetic mixing with ϵ < O(0.001) or Higgs-Abelian interactions with gλ−1/4 ≲ O(10−23).

In summary, the paper concludes that LISA observations offer strong prospects for constraining ultralight scalar and vector bosons through black hole spin measurements, with sensitivity spanning several orders of magnitude in boson mass, while direct detection through post-merger follow-up searches is considerably more challenging, particularly when accounting for the backreaction of superradiance on the spins of binary constituents.

Improvements to AI Systems Based on This Paper:

1. Physics-Aware Bayesian Inference for Excluded Parameter Spaces

• Improvement: Integrate the paper’s spin-down timescale (TS and 0.01 TS), boson mass ranges, and SNR thresholds into an AI-driven Bayesian model that automatically updates exclusion probabilities for scalar/vector bosons as new LISA mock catalogs are ingested.

• Capability: The AI can now produce real-time, posterior-driven exclusion maps (e.g., mass vs. spin) with confidence levels >0.99, directly replicating the paper’s methodology without manual recalibration.

2. Multi-Population Synthetic Catalog Generator with Superradiance Feedback

• Improvement: Train a generative AI on the three population models (PopIII, Q3delays, Q3nodelays) to synthesize realistic massive black hole binary catalogs, including pre-merger spin-down due to superradiance (as in the paper’s detection scenario).

• Capability: The AI can simulate thousands of LISA-like observations, predicting detection probabilities for follow-up signals (e.g., >80% for Q3nodelays in [2×10−16, 1.5×10−15] eV) and optimizing mission design (e.g., observation time allocation) under different astrophysical priors.

3. Automated Signal-to-Noise Ratio (SNR) Forecasting and Target Selection

• Improvement: Embed the lisabeta/IMRPhenomHM waveform models and SNR thresholds (≥20 for binaries, ≥10 for follow-ups) into a reinforcement learning agent that prioritizes which merger remnants to observe for quasi-monochromatic boson cloud signals.

• Capability: The AI can autonomously rank candidate remnants by expected SNR, accounting for spin-down backreaction, and dynamically adjust follow-up observation schedules to maximize detection probability within a 4-year LISA mission.

4. Cross-Sector Coupling Constraint Inference

• Improvement: Use the paper’s derived limits (e.g., quartic self-interaction f ≳ 1015–1016 GeV, kinetic mixing ϵ < 0.001, Higgs-Abelian gλ−1/4 ≲ 10−23) to train a classifier that flags which ultralight boson models (scalar vs. vector, with self-interactions or dark photon couplings) are consistent with LISA spin measurements.

• Capability: The AI can rapidly screen theoretical particle physics models, identifying which parameter combinations survive LISA constraints and which are excluded, accelerating model-building beyond gravitational-only analyses.

5. Uncertainty-Aware Extrapolation to Unobserved Mass Ranges

• Improvement: Implement a Gaussian process or neural network that learns the relationship between boson mass, spin-down timescale, and exclusion probability from the paper’s discrete mass ranges, then extrapolates to intermediate or higher masses with quantified uncertainty.

• Capability: The AI can predict LISA’s sensitivity for boson masses not explicitly simulated (e.g., 10−15–10−14 eV scalars), providing actionable guidance for future detector designs or joint analyses with other observatories (e.g., Einstein Telescope).

6. Backreaction-Aware Merger Remnant Property Prediction

• Improvement: Train a surrogate model on the paper’s fitting formulae that recompute remnant spins/masses after pre-merger superradiant spin-down, including the suppression of instability growth.

• Capability: The AI can quickly estimate final black hole properties for arbitrary initial spins and boson masses, enabling fast Monte Carlo simulations of LISA event rates and detection probabilities without expensive numerical relativity or SuperRad runs.

7. Automated Literature Consistency Checking

• Improvement: Build an AI that compares new LISA constraints (from this paper) against prior results (e.g., Brito et al. 2017) and flags discrepancies or confirmations, using natural language processing on abstracts and conclusions.

• Capability: The AI can maintain a live, self-updating knowledge base of ultralight boson constraints, alerting researchers when new observations tighten or relax existing exclusion regions, and suggesting reconciliations (e.g., catalog differences).