Spectral Sirens in Cosmology

Updated 28 May 2026

Spectral sirens are cosmological probes that use intrinsic spectral features in gravitational-wave sources, such as mass peaks and gaps, to deduce redshift and distance.
They employ hierarchical Bayesian inference and nonparametric models to map detector-frame masses to source-frame distributions, enabling precise estimations of H0 and related parameters.
Extensions including spinning, multi-spectral, and bosonic sirens enhance robustness by isolating sub-populations and leveraging unique spin or scalar features to counteract astrophysical systematics.

Spectral sirens are a class of cosmological and fundamental-physics observables in which spectral features—typically in gravitational-wave (GW) source populations, but more recently also scalar bosonic radiation from black hole (BH) superradiance—encode redshift and distance information without needing external electromagnetic (EM) counterparts or galaxy redshifts. The “spectral siren” methodology leverages intrinsic features in the source-frame distribution of a population (such as characteristic mass scales, gaps, or transitions) to statistically break the parameter degeneracies between intrinsic properties and cosmological expansion, allowing precise inference of the luminosity distance–redshift relation and cosmological parameters such as the Hubble constant $H_0$ purely from GW (or, prospectively, scalar) observations.

1. Conceptual Framework and Historical Context

Spectral sirens arose to address key limitations in GW cosmology. While bright standard sirens (compact-object mergers with EM counterparts) yield direct host redshifts, and dark sirens use statistical association with host galaxy catalogs, these methods become impractical for BBH-dominated catalogs, due to EM faintness and galaxy-incompleteness. Spectral sirens exploit the one-to-one mapping between detector-frame and source-frame masses, $m^\mathrm{det} = (1+z)m^\mathrm{src}$ , to infer cosmological redshift from population-level spectral features—such as peaks, bumps, dips, and roll-offs—in the intrinsic mass (or spin-mass) distribution (Hernandez et al., 2024, Mali et al., 2024, Hernandez et al., 3 Sep 2025, Bertheas et al., 6 Mar 2026, Tagliazucchi et al., 6 Jan 2026).

In this context, "spectral siren cosmology" refers to GW-only inference schemes that obtain cosmological constraints by comparing the observed (detector-frame) mass distribution to the source-frame distribution under varying cosmological parameters. This methodology has been extensively developed and validated on LIGO/Virgo-KAGRA catalogs and simulated datasets, and has recently been generalized to:

Nonparametric inference using Gaussian processes and B-splines (Hernandez et al., 2024, Tagliazucchi et al., 6 Jan 2026),
Spin-mass–correlated features ("spinning spectral sirens") (Tong et al., 15 Feb 2025),
Mixture-population (“multi-spectral sirens”) frameworks (Li et al., 2024),
Bosonic (scalar) spectral sirens via black hole superradiance (Gavilan-Martin et al., 26 Feb 2026).

2. Methodological Formalism

The central inference scheme for spectral sirens is a hierarchical Bayesian likelihood that links the observed population to both population features and cosmological expansion: $\mathcal{P}(\theta \mid \{d_i\}, N^\mathrm{G}) \propto \Pi(\theta) \mathcal{L}(\{d_i\}, N^\mathrm{G} \mid \theta)$ where $\theta$ includes cosmology (e.g., $H_0$ , $\Omega_m$ ), rate evolution, and population parameters (e.g., mass-distribution hyperparameters) (Cheng et al., 13 Mar 2026). The likelihood incorporates Poisson statistics for the expected number of detections, selection effects, and standard or nonparametric population priors (e.g., Gaussian processes over mass bins (Hernandez et al., 2024); adaptive B-splines (Tagliazucchi et al., 6 Jan 2026)).

Redshift is extracted via source-frame mass spectral features: observed detector-frame masses $m^\mathrm{det}$ are related to source-frame masses through the distance–redshift relation $D_L(z; H_0, \Omega_m)$ . Features in $p(m_\mathrm{src})$ become "stretched" in the $m^\mathrm{det}$ – $m^\mathrm{det} = (1+z)m^\mathrm{src}$ 0 plane under cosmological redshift, with their mapping sensitive to $m^\mathrm{det} = (1+z)m^\mathrm{src}$ 1 and other parameters (Hernandez et al., 3 Sep 2025).

Multi-spectral sirens generalize this to multiple sub-populations, allowing for sharper, less-confounded features by treating, for example, low-spin and high-spin BBH formation channels as distinct mixture components with separate mass/spin models and rate evolutions (Li et al., 2024).

Additional model ingredients include:

Population rate evolution $m^\mathrm{det} = (1+z)m^\mathrm{src}$ 2 (e.g., Madau–Dickinson or power-law forms),
Selection functions $m^\mathrm{det} = (1+z)m^\mathrm{src}$ 3 encoding detector sensitivity and observation cuts,
Explicit modeling of redshift evolution in population features to capture possible biases (Pierra et al., 2023).

For spinning spectral sirens, the joint mass–spin distribution is modeled; spin parameters (e.g., $m^\mathrm{det} = (1+z)m^\mathrm{src}$ 4) are not redshifted and can identify robust mass scales immune to evolutionary bias (Tong et al., 15 Feb 2025).

3. Population Features and Information Content

Spectral sirens derive their constraining power from the presence of sharp or semi-sharp features in the source-frame distribution. Empirical analyses of LIGO/Virgo data reveal multiple such features:

Peaks at $m^\mathrm{det} = (1+z)m^\mathrm{src}$ 5 and $m^\mathrm{det} = (1+z)m^\mathrm{src}$ 6,
A high-mass roll-off above $m^\mathrm{det} = (1+z)m^\mathrm{src}$ 7 (Mali et al., 2024, Tagliazucchi et al., 6 Jan 2026).

Correlation studies show that the high-mass peak near $m^\mathrm{det} = (1+z)m^\mathrm{src}$ 8 dominates cosmological sensitivity (correlation coefficient $m^\mathrm{det} = (1+z)m^\mathrm{src}$ 9 with $\mathcal{P}(\theta \mid \{d_i\}, N^\mathrm{G}) \propto \Pi(\theta) \mathcal{L}(\{d_i\}, N^\mathrm{G} \mid \theta)$ 0), with the low-mass peak and roll-off also supplying information. Nonparametric summary statistics on these intervals (e.g., windowed means and standard deviations) correlate independently with $\mathcal{P}(\theta \mid \{d_i\}, N^\mathrm{G}) \propto \Pi(\theta) \mathcal{L}(\{d_i\}, N^\mathrm{G} \mid \theta)$ 1, enabling checks on astrophysical systematics. Multiple features allow cross-validation against evolutionary bias; conditionally, their correlations with $\mathcal{P}(\theta \mid \{d_i\}, N^\mathrm{G}) \propto \Pi(\theta) \mathcal{L}(\{d_i\}, N^\mathrm{G} \mid \theta)$ 2 are almost independent (Mali et al., 2024).

Higher-fidelity mass models—either flexible parametric forms (combinations of power laws and Gaussians (Hernandez et al., 3 Sep 2025, Bertheas et al., 6 Mar 2026)) or data-driven semiparametric models—better exploit these features, improving $\mathcal{P}(\theta \mid \{d_i\}, N^\mathrm{G}) \propto \Pi(\theta) \mathcal{L}(\{d_i\}, N^\mathrm{G} \mid \theta)$ 3 precision by 12–21% relative to simple models (Tagliazucchi et al., 6 Jan 2026).

4. Systematics, Evolution, and Self-Calibration

The primary sources of systematic error in spectral siren cosmology are:

Incorrect or insufficiently flexible modeling of the source mass/spin distribution,
Neglect of potential redshift evolution of mass features (e.g., drift of population peaks with $\mathcal{P}(\theta \mid \{d_i\}, N^\mathrm{G}) \propto \Pi(\theta) \mathcal{L}(\{d_i\}, N^\mathrm{G} \mid \theta)$ 4),
Implicit astrophysical assumptions on population composition or selection (Pierra et al., 2023, Bertheas et al., 6 Mar 2026).

Simulations demonstrate that applying a redshift-independent mass model to an evolving population causes percent-level to $\mathcal{P}(\theta \mid \{d_i\}, N^\mathrm{G}) \propto \Pi(\theta) \mathcal{L}(\{d_i\}, N^\mathrm{G} \mid \theta)$ 5-level biases in $\mathcal{P}(\theta \mid \{d_i\}, N^\mathrm{G}) \propto \Pi(\theta) \mathcal{L}(\{d_i\}, N^\mathrm{G} \mid \theta)$ 6 for large catalogs (Pierra et al., 2023). Realistic multi-channel (e.g., field/dynamical) evolution can produce mis-estimated source-frame features if the model cannot adapt to topology changes.

Spinning spectral sirens introduce a self-calibrating check: spin–mass transitions (e.g., at the edge of the pair-instability gap, where BH effective spins change abruptly) are not redshifted and thus provide an astrophysically robust “ruler.” These can isolate cosmological redshift even if other features drift (Tong et al., 15 Feb 2025). Empirical evidence from GWTC-3 supports the stability of the high-spin transition at $\mathcal{P}(\theta \mid \{d_i\}, N^\mathrm{G}) \propto \Pi(\theta) \mathcal{L}(\{d_i\}, N^\mathrm{G} \mid \theta)$ 7.

5. Extensions: Multi-Spectral and Bosonic Sirens

The multi-spectral siren paradigm separates the BBH (or compact-object) population into subgroups—distinguished by formation channel (field vs. cluster) or by spin, for example—with each subpopulation having a distinct mass/spin model and selection function. Simulations and data analyses show that multi-spectral modeling sharpens spectral features and yields 19–26% improvements in $\mathcal{P}(\theta \mid \{d_i\}, N^\mathrm{G}) \propto \Pi(\theta) \mathcal{L}(\{d_i\}, N^\mathrm{G} \mid \theta)$ 8 constraints, as features otherwise blurred by superposition become clearer (Li et al., 2024).

A distinct extension is the notion of BH scalar sirens. In these scenarios, light bosonic fields (e.g., axion-like particles) extract angular momentum from spinning black holes via superradiant instabilities, populating bosonic clouds which emit narrow or broad scalar radiation. Under suitable scalar self-coupling, long-lived "scalar sirens" are produced, emitting non-relativistic scalars with broad spectra up to $\mathcal{P}(\theta \mid \{d_i\}, N^\mathrm{G}) \propto \Pi(\theta) \mathcal{L}(\{d_i\}, N^\mathrm{G} \mid \theta)$ 9 (Gavilan-Martin et al., 26 Feb 2026). These scalar backgrounds encode the mass and spin distribution of the underlying BH population and offer a novel window for probing both ultralight bosons and the cosmological BH abundance.

BH scalar sirens, detected through interactions with terrestrial sensors (spin-precession or axion-photon mixing), probe boson mass/frequency bands independent of standard cosmological or early-universe production and allow for constraints on otherwise invisible BHs. The expected signal is up to two orders of magnitude larger than from misaligned cosmic scalars in the relevant mass range, with distinct velocity and spectral signatures.

6. Practical Performance: Forecasts, Pipelines, and Future Prospects

Large-scale blinded mock challenges, using simulated third-generation GW detector data, have validated the scalability and precision of spectral siren pipelines (ICAROGW, CHIMERA, pymcpop-gw). With $\theta$ 0 events per year, current GPU-accelerated pipelines realize percent-level precision at $\theta$ 1, and $\theta$ 2 precision at the $\theta$ 310% level, improving as $\theta$ 4 (Tagliazucchi et al., 19 Feb 2026). The dominant contributors to the cosmological information are low-distance events near sharp mass features and high-mass events at larger distances for $\theta$ 5.

Observationally, current catalogs (e.g., GWTC-4) combined with a single bright siren (GW170817) yield $\theta$ 6 posteriors with 10–17% precision using spectral siren methods—already comparable with early dark-siren methods and robust to different mass models (Hernandez et al., 3 Sep 2025). Further statistical gains are expected as BBH catalogs grow; thousands of events (as expected from Einstein Telescope and Cosmic Explorer) will achieve subpercent precision on $\theta$ 7 and broad constraints on modified gravity or dark energy (Tagliazucchi et al., 19 Feb 2026, Borghi et al., 20 Dec 2025). Such precision is contingent on continued improvements in population model flexibility and treatment of evolutionary effects.

For bosonic sirens, the integrated Milky Way background, considering $\theta$ 8 isolated stellar-mass BHs, constitutes a measurable scalar wind with distinct spectral and velocity properties, motivating both broadband and resonant terrestrial searches (Gavilan-Martin et al., 26 Feb 2026).

7. Outlook and Complementarity

Spectral sirens constitute a critical complementary avenue in the GW cosmology program, orthogonal to both the EM distance ladder and galaxy-catalog dark sirens. Their robustness to EM incompleteness and potential for self-calibration against astrophysical evolution herald a path to percent-level cosmology with GW-only data. Further, the extension of the spectral-siren paradigm to scalar emissions from BH superradiance broadens the scope of spectral observables beyond GWs, tying together fundamental physics, stellar evolution, and precision cosmology.

Key remaining challenges are flexible population modeling (including nonparametric joint distributions, full mass–redshift–spin correlation structures), principled incorporation of redshift evolution, and the exploitation of new and multiple spectral features to guard against astrophysical bias. The synergy among spectral, spinning, multi-spectral, and bosonic sirens, combined with anticipated third-generation GW detector yields, positions spectral sirens as a principal probe of the expansion and composition of the Universe in the coming decade.

References:

(Hernandez et al., 2024, Mali et al., 2024, Tong et al., 15 Feb 2025, Hernandez et al., 3 Sep 2025, Tagliazucchi et al., 6 Jan 2026, Gavilan-Martin et al., 26 Feb 2026, Tagliazucchi et al., 19 Feb 2026, Bertheas et al., 6 Mar 2026, Cheng et al., 13 Mar 2026, Pierra et al., 2023, Li et al., 2024, Borghi et al., 20 Dec 2025)