Mind the peak: improving cosmological constraints from GWTC-4.0 spectral sirens using semiparametric mass models

By applying a novel semiparametric B-spline model to 137 binary black hole events from GWTC-4.0, this study resolves three distinct mass distribution peaks and achieves a 12–21% improvement in the precision of the Hubble constant ($H_0$) compared to standard parametric models, demonstrating that capturing the full complexity of the mass distribution is essential for maximizing the cosmological potential of gravitational wave spectral sirens.

The Gist

Imagine the universe is a giant, expanding balloon. For decades, scientists have been trying to measure exactly how fast this balloon is inflating (a rate called the Hubble Constant, or $H_0$). Usually, they do this by looking at light from distant stars, but there's a disagreement between different measurement methods.

Enter Gravitational Waves. These are ripples in spacetime caused by massive objects crashing together, like two black holes merging. These events act as "standard sirens"—like a lighthouse in the dark. If we know how loud the siren should be (based on the physics of the black holes) and how loud it actually sounds to us, we can calculate how far away it is.

However, there's a catch: The "loudness" of a black hole merger depends on its mass. But because the universe is expanding, the mass we measure looks different from the mass the black holes actually had when they were born. This creates a confusing mix-up, or "degeneracy," where we can't easily tell if an object is heavy and close, or light and far away.

The Problem: Guessing the Shape of the Black Hole "Family"

To solve this mix-up, scientists use a trick called Spectral Sirens. They look at the entire population of black holes. If you know the general shape of the "family tree" of black hole masses (how many are small, how many are huge, and where the common sizes are), you can untangle the distance and mass confusion.

For a long time, scientists have tried to guess the shape of this family tree using simple mathematical formulas (like a straight line with a few bumps). The authors of this paper argue that these simple guesses are too rigid. They are like trying to describe a complex mountain range using only a few flat triangles. You miss the valleys, the sharp peaks, and the hidden ridges.

The Solution: A Flexible, "Smart" Map

The team, led by Matteo Tagliazucchi, decided to stop guessing the shape and instead let the data draw the map for them. They used a new method called a semiparametric model based on something called Bsplines.

Think of it this way:

• Old Method (Parametric): Imagine trying to draw a coastline using only a ruler and a protractor. You can only make straight lines and perfect circles. It's easy, but it doesn't look like the real coast.
• New Method (Semiparametric): Imagine drawing that same coastline with a flexible, bendable wire. You can bend the wire to match every tiny inlet and jagged rock, but you only bend it where the data tells you to.

They analyzed 137 black hole mergers from the latest catalog (GWTC-4.0). Instead of forcing the data into a pre-made shape, their "flexible wire" model automatically found the most important spots to bend.

What They Found

By letting the model be flexible, they discovered that the black hole mass distribution isn't just a couple of smooth bumps. It has three distinct peaks (hills) at specific masses:

1. Around 10 times the mass of our Sun.
2. Around 18 times the mass of our Sun.
3. Around 33 times the mass of our Sun.

The old, rigid models missed the middle peak (18 solar masses) and smoothed over the others. The new model saw them clearly.

Why This Matters for the Universe

Here is the magic part: The exact position of these "hills" in the black hole family tree is tightly linked to how fast the universe is expanding ($H_0$).

Because the new model captured these three hills accurately, it could untangle the distance-mass confusion much better than the old models.

• The Result: Their measurement of the universe's expansion rate became 12% to 21% more precise than previous attempts using rigid models.
• The Number: They calculated the expansion rate to be roughly 57.8 km/s/Mpc (with a margin of error).

The Takeaway

The paper concludes that to get the best possible answer about how the universe is expanding, we cannot rely on simple, pre-set guesses about what black holes look like. We need to use flexible, data-driven tools that can "feel" the subtle bumps and peaks in the data.

Just as a high-resolution map reveals hidden paths that a sketch misses, this new flexible model reveals hidden structures in the black hole population, allowing us to measure the cosmos with greater clarity. The authors emphasize that as we find more black holes in the future, capturing these full details will be essential for turning gravitational waves into a precise ruler for the universe.

Technical Summary

1. Problem Statement

Gravitational wave (GW) "spectral sirens" offer a method to measure the Hubble constant ($H_0$) by statistically breaking the mass-redshift degeneracy using the intrinsic mass distribution (MD) of binary black holes (BBHs). However, the precision and accuracy of these constraints rely heavily on the fidelity of the MD model.

• Limitation of Current Approaches: Existing analyses (e.g., GWTC-3 and early GWTC-4) rely on parametric models (e.g., power laws with fixed Gaussian peaks). These models assume a specific functional form for the MD. If the assumed form is incorrect or oversimplified (e.g., missing substructures), it introduces biases in the inferred $H_0$ and reduces the method's constraining power.
• The Challenge: Guessing the correct MD shape in advance is difficult. As catalogs grow (GWTC-4.0), the data reveals more complex substructures that rigid parametric templates may fail to capture.

2. Methodology

The authors analyzed 137 BBH events from the GWTC-4.0 catalog (O1–O4a) using a hierarchical Bayesian framework within the CHIMERA pipeline. They compared a standard parametric model against a novel semiparametric approach.

A. The Semiparametric Model (Bspline)

Instead of assuming a fixed number of Gaussian peaks, the authors modeled the primary mass distribution $p(m_1)$ using a Bspline expansion:
$$p(m_1 | \lambda_m) \propto SP(m_1) \cdot \exp(s^{(d)}(m_1))$$

• Base: A truncated power law ($SP$) with a low-edge smoothing factor.
• Flexibility: A Bspline term $s^{(d)}(m_1)$ composed of basis functions and coefficients $\{c_i\}$ that allow the model to adaptively deviate from the power law to fit data features.
• Knot Placement Strategy: A critical innovation is the data-driven optimization of knot positions.
  • Rather than using uniformly or logarithmically spaced knots, the authors computed the mean observed source-frame mass distribution for various $H_0$ values.
  • Knot positions were identified as the points of highest variation (peaks of the derivative of the log-distribution).
  • A clustering algorithm (Gaussian Mixture Models) was used to determine the optimal number and location of knots, penalizing complexity via the Bayesian Information Criterion (BIC) to prevent overfitting.

B. Comparison Models

• Parametric Baseline (pl2p): A Power Law + Double Peak model (truncated power law + two Gaussians), consistent with recent LVK analyses.
• Semiparametric Variants: Models with varying numbers of knots (10, 12, 14, 16) and different prior widths ($\sigma$) on the spline coefficients.

3. Key Contributions

1. Novel Data-Driven Knot Placement: Introduced a method to automatically place Bspline knots around the most informative structures in the mass distribution, rather than relying on arbitrary spacing.
2. Discovery of Three Distinct Peaks: The flexible semiparametric model resolved three distinct peaks in the BBH mass distribution at approximately 10, 18, and 33 $M_\odot$. Standard parametric models only captured two broader peaks, effectively smoothing out the 18 $M_\odot$ feature.
3. Quantification of Cosmological Impact: Demonstrated that capturing these substructures directly translates to tighter constraints on $H_0$.

4. Results

A. Model Comparison

• Bayes Factors (BF): The semiparametric model with 14 data-driven knots and a prior width $\sigma=2$ (pls-dd-14-G2) was strongly preferred over the parametric baseline, with a Bayes factor of 226.
• DIC: The most flexible model (pls-dd-14-G5) showed the lowest Deviance Information Criterion, indicating the best fit.
• Logarithmic Spacing: Models with logarithmically spaced knots performed significantly worse (BF = 7.28), proving that where the knots are placed is as important as the number of knots.

B. Mass Distribution Constraints

• The semiparametric models successfully recovered the substructure at 18 $M_\odot$, which was missed by the parametric model.
• A potential small bump at $\sim 60 M_\odot$ was hinted at with higher knot counts, consistent with other recent findings.

C. Cosmological Constraints ($H_0$)

Capturing the full complexity of the MD significantly improved the precision of $H_0$:

• Best Case (pls-dd-14-G5): $H_0 = 61.7^{+19.3}_{-14.9}$ km/s/Mpc. This represents a ~21% improvement in precision compared to the parametric model.
• BF-Favored Case (pls-dd-14-G2): $H_0 = 57.8^{+21.9}_{-20.6}$ km/s/Mpc. This represents a ~12% improvement.
• Correlation Analysis: The study identified that events near the newly resolved 18 $M_\odot$ peak carry significant cosmological information. In the semiparametric model, events in the detector-frame mass range [20, 35] $M_\odot$ showed a strong anticorrelation with $H_0$, a pattern obscured in the parametric model.

5. Significance

• Validation of Spectral Sirens: The results confirm that the "spectral siren" method is highly sensitive to the accuracy of the mass distribution model. Incorrect assumptions lead to suboptimal constraints.
• Future-Proofing: As GW catalogs grow (GWTC-5 and beyond), the mass distribution will likely reveal even more complex substructures. The proposed semiparametric, data-driven approach provides a robust framework to adapt to these complexities without manual re-tuning of model parameters.
• Cosmological Tension: By reducing biases and improving precision, this method enhances the potential of GWs to resolve the current tension between local and early-Universe measurements of the Hubble constant.

Conclusion: The paper demonstrates that moving from rigid parametric templates to flexible, data-driven semiparametric models is essential for maximizing the cosmological potential of gravitational wave observations.