The Missing Multipole Problem: Investigating biases from model starting frequency in gravitational-wave analyses

This paper demonstrates that using gravitational-wave waveform models with inappropriate starting frequencies can lead to biased source property inference for high-mass, asymmetric binary black hole systems, particularly when signal-to-noise ratios are high enough for systematic errors to dominate statistical uncertainties.

The Gist

The Big Picture: Listening to the Cosmic Symphony

Imagine the universe is a giant concert hall, and when two black holes collide, they create a cosmic symphony. Scientists use gravitational wave detectors (like giant, ultra-sensitive ears) to listen to this music.

To understand what happened in the collision—how heavy the black holes were, how fast they were spinning, and how far away they are—scientists compare the sound they hear against a library of theoretical songs (called "templates"). If the real sound matches a theoretical song perfectly, they know the properties of the black holes.

The Problem: The "Missing Notes"

This paper discovers a hidden flaw in how some of these theoretical songs are written.

Think of a black hole collision like a musical chord.

• The main note (the loudest, most obvious sound) is the "quadrupole."
• But there are also higher harmonics (fainter, more complex notes) called "multipoles." These are like the high-pitched violin notes or the deep bass undertones that add richness to the chord.

The Issue:
When scientists generate these theoretical songs using computer models, they often have to decide when to start the song.

• If they start the song too late (say, at 20 Hz, which is a low hum), they accidentally cut off the beginning of the fainter, higher harmonics.
• It's like trying to identify a song by only listening to the chorus, ignoring the verses and the intro. You might think the song is about love, but if you missed the first verse, you might think it's about heartbreak.

In the world of black holes, missing these "intro notes" (the low-frequency power of the higher multipoles) makes the computer think the black holes are different than they actually are.

The Investigation: Tuning the Radio

The authors of this paper decided to test how much this "cutting off" matters. They focused on Intermediate Mass Black Holes (IMBHs).

• Analogy: Think of stellar-mass black holes as small, lightweight guitars. IMBHs are like massive, heavy double-basses. Because they are so heavy, they crash together very quickly, and their "song" is short and intense. Because the song is so short, every single note counts. If you miss the first few seconds, you miss the whole story.

They ran thousands of simulations, essentially playing the "real" song (with all notes) and then trying to match it against three different versions of the theoretical song:

1. The Full Version: Starts early (10 Hz), capturing every note, including the faint harmonics.
2. The "Good Enough" Version: Starts a bit later (13 Hz), missing the very faintest notes.
3. The "Cut Short" Version: Starts late (20 Hz), missing the significant high-pitched harmonics.

The Findings: When Does It Matter?

The results were surprising and depended on three things: How loud the signal is, how heavy the black holes are, and how lopsided the pair is.

1. The "Loudness" Factor (Signal-to-Noise Ratio)

• Quiet Signals (Low Volume): If the signal is faint (like hearing a whisper in a noisy room), the "static" (statistical noise) is so loud that it doesn't matter if you miss a few notes. The error from the noise is bigger than the error from missing the notes.
  • Rule of Thumb: If the signal is quiet (SNR < 20), you can safely start the song late (20 Hz).
• Loud Signals (High Volume): If the signal is loud and clear (SNR > 20), the "static" is quiet. Now, missing those extra notes becomes a huge problem. The computer gets confused and gives you the wrong answer.
  • Rule of Thumb: If the signal is loud, you must start the song early to catch all the notes.

2. The "Weight" Factor (Total Mass)

• Lighter IMBHs: If the black holes aren't too heavy, missing the extra notes causes a moderate error.
• Heavier IMBHs: As the black holes get heavier (over 300 times the mass of our Sun), the "song" gets shorter and the missing notes become critical. If you start the song late, the computer might think the black holes are much lighter or heavier than they really are.

3. The "Shape" Factor (Mass Ratio)

• If the two black holes are the same size (a perfect duet), missing the notes is bad.
• If they are very different sizes (a heavy bass and a light guitar), missing the notes is terrible. The computer gets completely confused about the size of the smaller black hole.

The Real-World Test: GW231123

The authors didn't just play with fake data; they tested this on a real event detected recently, called GW231123.

• When they analyzed this real event using the "Cut Short" version (starting at 20 Hz), the results were weird and "bimodal" (the computer couldn't decide if the black holes were heavy or light).
• When they used the "Full Version" (starting at 10 Hz), the results were clear and consistent.
• Conclusion: For this real event, the "Missing Multipole Problem" was actually distorting our view of the universe.

The Takeaway: How to Listen Correctly

The paper gives a simple guide for future gravitational wave hunters:

1. If the signal is faint: You can be lazy and start your analysis at 20 Hz. The noise will hide the errors anyway.
2. If the signal is loud (but not super loud): You must start at 13 Hz to catch the "3rd note" (the 3,3 multipole).
3. If the signal is very loud, the black holes are very heavy, or they are very different sizes: You must start at 10 Hz to catch all the notes, including the "4th note" (the 4,4 multipole).

In summary: To accurately weigh the heaviest, most dramatic black hole collisions in the universe, we cannot just listen to the main melody. We must tune our radio to catch the faint, high-pitched harmonics right from the very beginning of the song, or else we will misjudge the stars entirely.

Technical Summary

1. Problem Statement

The paper addresses a systematic bias in Gravitational Wave (GW) parameter estimation known as the "Missing Multipole Problem."

• Context: GW signals are decomposed into spin-weighted spherical harmonics (multipoles). While the dominant quadrupole mode $(\ell, |m|) = (2, 2)$ is the strongest, higher-order multipoles (e.g., $(3, 3)$, $(4, 4)$) become significant in systems with asymmetric mass ratios, high total masses, and specific orbital inclinations.
• The Issue: Time-domain waveform models (e.g., NRSur7dq4) are converted to the frequency domain for Bayesian analysis. In time-domain models, higher-order multipoles start at different characteristic frequencies relative to the dominant mode. Specifically, $f_{\ell m} = \frac{m}{2} f_{22}$.
• The Bias: Standard GW analyses typically begin at a minimum frequency ($f_{\text{low}}$) of 20 Hz. If a time-domain template is started such that the dominant mode $(2, 2)$ begins at 20 Hz, higher-order modes like $(3, 3)$ and $(4, 4)$ do not enter the analysis band until 30 Hz and 40 Hz, respectively. This exclusion of low-frequency power from higher-order multipoles can lead to biased estimates of source properties (mass, spin, inclination), particularly for short-duration signals like Intermediate Mass Black Hole (IMBH) mergers.
• Previous Work: A prior study by Islam et al. (2023) concluded this effect was minor, but their analysis focused on equal-mass, moderate-mass systems where higher-order multipoles are less dominant.

2. Methodology

The authors conducted a systematic injection and recovery study using Bayesian inference to quantify these biases.

• Waveform Model: They used NRSur7dq4, a numerical relativity surrogate model that includes spin precession and multipoles up to $\ell_{\text{max}} = m_{\text{max}} = 4$.
• Simulation Setup:
  • Signals: Injected synthetic GW signals into "zero-noise" to isolate systematic biases from statistical noise.
  • Parameter Space: Focused on light IMBH binaries with total masses $M \in [200, 450] M_\odot$, mass ratios $q \in [0.25, 1]$, and varying spins and inclinations.
  • Detector Network: Simulated a network of two Advanced LIGO and one Advanced Virgo detectors at design sensitivity.
• Analysis Variables:
  • Injected Signal: Always generated with the dominant mode starting at $f_{22} = 10$ Hz (ensuring all multipoles are present from 20 Hz).
  • Template Starting Frequencies: Three separate analyses were run for each injection using templates starting at:
    1. $f_{22} = 10$ Hz (Baseline/Unbiased: All multipoles present from 20 Hz).
    2. $f_{22} = 13$ Hz (Missing $(4, 4)$ mode below 40 Hz).
    3. $f_{22} = 20$ Hz (Missing $(3, 3)$ and $(4, 4)$ modes below 30 Hz and 40 Hz).
• Inference Tool: Bayesian inference performed using Bilby with the dynesty nested sampling algorithm.
• Metrics:
  • Mahalanobis Recovery Score ($r_M$): Measures the fraction of posterior realizations where the 90% credible interval contains the true injected value.
  • Bayes Factors ($B$): Used to compare the evidence for different starting frequencies (e.g., $f_{22}=10$ Hz vs. $f_{22}=20$ Hz).

3. Key Contributions

• Quantification of Bias: The paper provides a rigorous quantification of how template starting frequency biases parameter estimation, specifically identifying regimes where the "missing multipole" problem is critical.
• SNR and Mass Thresholds: It establishes specific thresholds for Signal-to-Noise Ratio (SNR) and total mass where higher-order multipoles become essential for unbiased recovery.
• Real-World Application: The study validates its findings by re-analyzing the real GW event GW231123, demonstrating that the choice of starting frequency alters the inferred source properties even for real data.
• Correction of Previous Literature: It challenges the conclusions of Islam et al. (2023), showing that biases are significant for IMBH systems and high-SNR detections, contrary to previous assumptions that they were negligible.

4. Key Results

The study reveals that the impact of missing multipoles depends heavily on the system's total mass, mass ratio, and the SNR of the detection.

• Total Mass Dependence:
  • $M \lesssim 250 M_\odot$: Biases are negligible; all starting frequencies recover parameters accurately.
  • $M \gtrsim 300 M_\odot$: Significant biases appear. Templates starting at 20 Hz underestimate total mass and overestimate mass asymmetry.
  • $M \gtrsim 450 M_\odot$: Even starting at 13 Hz (missing the $(4, 4)$ mode) introduces significant bias.
• Signal-to-Noise Ratio (SNR) Dependence:
  • $\rho < 20$: Statistical uncertainties dominate; starting at 20 Hz is safe.
  • $20 \lesssim \rho \lesssim 70$: The $(3, 3)$ multipole becomes critical. Templates must start at 13 Hz or lower to avoid bias.
  • $\rho \gtrsim 70$: The $(4, 4)$ multipole becomes critical. Templates must start at 10 Hz or lower.
• Mass Ratio and Inclination:
  • Biases are most severe for equal-mass systems (due to shorter inspirals) and highly asymmetric systems (where higher multipoles are intrinsically stronger).
  • Edge-on and face-on systems show different sensitivities, but the $(3, 3)$ mode is generally required for accurate recovery across all orientations for high-mass systems.
• GW231123 Case Study:
  • Re-analyzing GW231123 ($\rho \approx 22.6$) showed that using a 20 Hz start frequency resulted in a bimodal posterior for total mass with a preference for asymmetric masses, whereas 10 Hz and 13 Hz starts yielded consistent results. The Bayes factor favored the 10 Hz start over the 20 Hz start ($\log_{10} B = 2.2$).

5. Significance and Recommendations

The paper provides critical guidelines for current and future GW observations, particularly for the search for Intermediate Mass Black Holes (IMBHs) and events in the pair-instability mass gap.

Recommended Starting Frequencies ($f_{22}$) based on SNR ($\rho$):

1. $\rho < 20$: Start at 20 Hz. (Statistical noise dominates; missing multipoles are irrelevant).
2. $20 \leq \rho \leq 70$: Start at 13 Hz (or lower). This ensures the $(3, 3)$ multipole is included, preventing bias in mass and spin estimates.
3. $\rho > 70$: Start at 10 Hz (or lower). This ensures the $(4, 4)$ multipole is included, which is essential for high-mass ($>250 M_\odot$) and high-SNR systems.

Implications:

• Astrophysical Interpretation: Incorrect starting frequencies can lead to erroneous conclusions about the formation channels of black holes (e.g., whether they reside in the pair-instability mass gap).
• Future Detectors: As next-generation detectors (e.g., Cosmic Explorer, Einstein Telescope) improve low-frequency sensitivity, the "missing multipole" problem will become even more pronounced, making these recommendations vital for future data analysis pipelines.
• Modeling: Time-domain models must be carefully managed during frequency-domain conversion to ensure all relevant multipoles are captured within the detector's sensitive band.