⚛️ general relativity

Impact of Correlated Test-Mass Noise on Stochastic Gravitational-Wave Background Searches with Space-Based Interferometers

This study demonstrates that neglecting correlations between test-mass acceleration noises in space-based interferometers can significantly bias stochastic gravitational-wave background parameter inference, whereas explicitly modeling these correlations in Bayesian templates ensures accurate recovery of the signal and noise parameters.

Original authors: Jing-yi Wu, Yong Tang

Published 2026-10-07

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Jing-yi Wu, Yong Tang

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Deep in the quiet hum of the cosmos, there exists a background roar of gravitational waves, a faint static left over from the birth of the universe and the chaotic collisions of distant stars. Unlike the sharp, distinct chirps of individual black holes merging, this background is a continuous, unresolvable fog of ripples in spacetime. To hear it, scientists are building massive laser interferometers in space, such as the planned LISA, Taiji, and TianQin missions. These instruments consist of three spacecraft flying in a giant triangle, using lasers to measure tiny changes in distance between them. The goal is to separate the whisper of this cosmic background from the much louder hiss of the instruments themselves. If the team cannot perfectly understand the noise their own machines make, they might mistake a glitch in the hardware for a signal from the early universe, leading to a false discovery or a wrong measurement of how the universe began.

The core challenge lies in the test masses, the free-floating mirrors inside each spacecraft that serve as the reference points for the laser measurements. These mirrors are incredibly sensitive, and their motion is disturbed by everything from magnetic fields to the pressure of residual gas. In a perfect world, the jitters of the mirrors in one spacecraft would be entirely independent of the mirrors in the neighboring spacecraft. However, evidence from a previous mission, LISA Pathfinder, suggests that these mirrors might actually "talk" to each other. Because they share the same spacecraft environment and control systems, a disturbance affecting one mirror could simultaneously nudge the other, creating a correlated noise that looks suspiciously like a real gravitational wave signal. If the data analysis software assumes these mirrors are independent when they are actually linked, the math will try to explain this extra, correlated noise by inventing a stronger or differently shaped gravitational wave background than what actually exists.

In this study, researchers Jing-yi Wu and Yong Tang simulated exactly this scenario to see how much it would mess up the results. They created a realistic digital model of the Taiji mission, filling it with a fake gravitational wave background and the expected instrumental noise. Crucially, they injected a specific type of correlated noise between the test masses, varying the strength of this connection from zero up to a level where the mirrors were moving in perfect lockstep. They then ran two different types of analysis on this simulated data. The first analysis was "honest" about the problem: it included a mathematical model that accounted for the possibility that the mirrors were correlated. The second analysis was "standard": it assumed the mirrors were completely independent, ignoring the correlation entirely.

The results were clear and concerning. When the researchers used the "standard" model on data that actually contained correlated noise, the software got confused. It tried to make sense of the extra noise by shifting the values of the gravitational wave parameters. Depending on the specific nature of the noise, the software either guessed the wrong shape for the cosmic background or the wrong strength of the instrument's own noise. As the correlation between the mirrors got stronger, these errors grew larger, systematically pushing the answers away from the truth. In contrast, the "honest" model, which knew about the correlation, recovered the correct values for the gravitational waves and the instrument noise with high accuracy, regardless of how strong the correlation was.

The study highlights that the direction of the error depends on the frequency of the noise. If the correlated noise is uniform across all frequencies, the analysis tends to mistake it for a change in the slope of the gravitational wave background. If the noise follows the same pattern as the instrument's own natural vibrations, the analysis confuses it with the strength of the instrument's noise. This means that simply knowing how strong the correlation is not enough; scientists must also understand how that correlation changes with frequency to avoid biasing their results. The work does not prove that this correlation will definitely ruin future missions, but it demonstrates through simulation that ignoring it is a dangerous gamble. It suggests that for space-based gravitational wave detectors to succeed, their noise models must be sophisticated enough to include these subtle, shared disturbances between test masses, ensuring that what they hear is truly the voice of the universe and not just the echo of their own machinery.

Technical Summary: Impact of Correlated Test-Mass Noise on Stochastic Gravitational-Wave Background Searches

Problem Statement
Future space-based laser interferometers (e.g., LISA, Taiji, TianQin) aim to detect the stochastic gravitational-wave background (SGWB), which encodes information about unresolved astrophysical populations and early-Universe cosmological processes. Unlike resolved compact-binary signals, the SGWB manifests as a statistical excess of power across a frequency range. Distinguishing this signal from instrumental noise is critical, as an incomplete noise model can be misinterpreted as part of the SGWB, biasing the inferred spectrum and its parameters.

Evidence from the LISA Pathfinder mission indicates that environmental and control variables can exhibit coherence with differential acceleration, suggesting that acceleration (ACC) noise between adjacent test masses (TMs) in a spacecraft may be correlated. In a future mission, adjacent TMs share subsystems (magnetic, thermal, residual-gas, and control-related disturbances), potentially creating correlated noise components. If these correlations exist in the data but are omitted from the inference model, the resulting mismatch may be absorbed by the parameters retained in the template, leading to biased inference of both the SGWB and instrumental noise characteristics.

Methodology
The authors employ a forward-modeling framework to quantify the bias introduced when correlated ACC noise is present in mock data but absent from the analysis template.

Signal and Noise Modeling:
- Single-Link Model: The study defines observables for a three-spacecraft constellation, incorporating gravitational wave (GW) induced phase fluctuations, test-mass jitter ( $\delta_{ij}$ ), and optical bench motion.
- Correlated Noise Prescription: Two phenomenological models for the cross-spectrum of ACC noise between adjacent TMs in the same spacecraft are introduced:
  - White-Noise (WN): A frequency-independent cross-spectrum scaled to the minimum of the auto-spectrum.
  - PSD-Scaled (PS): A cross-spectrum that follows the exact frequency dependence of the ACC power spectral density (PSD).
- Propagation: The model propagates SGWB, ACC, and optical metrology system (OMS) noise through the moving-orbit response into Time-Delay Interferometry (TDI) observables. The data is segmented into 2.5-day intervals to handle non-stationarity due to orbital evolution.
Inference Framework:
- Data: Mock data is generated using the Taiji Data Challenge II toolkit, containing injected SGWB (power-law spectrum), OMS noise, and ACC noise with varying correlation amplitudes ( $\rho$ ).
- Templates: Two Bayesian frequency-domain templates are compared using the A and E TDI auto-spectra:
  - Corr. Template: Includes the correlated-noise covariance ( $C_{acc}$ ) in the likelihood model.
  - Non-Corr. Template: Assumes ACC noises are uncorrelated (sets $C_{acc} = 0$ ).
- Analysis: Nested sampling is used to estimate posterior distributions for parameters including SGWB amplitude ( $\Omega_0$ ), spectral slope ( $\gamma$ ), OMS noise amplitude ( $s_{oms}$ ), ACC noise amplitude ( $s_{acc}$ ), and the correlation amplitude ( $\rho$ ) for the Corr. Template.

Key Results
The study isolates the impact of the missing cross-spectral structure by holding the signal response, instrumental auto-spectra, and analysis bands fixed.

Systematic Biases: When correlated ACC noise is present in the mock data, the Non-Corr. Template exhibits systematic parameter offsets that generally increase with the correlation amplitude ( $\rho$ ). The Corr. Template recovers the injected SGWB and instrumental-noise parameters substantially more accurately.
Degeneracy Dependence: The specific parameter that absorbs the mismatch depends on the spectral shape of the correlation:
- For the WN model, the correlated component is most strongly degenerate with the SGWB spectral slope ( $\gamma$ ).
- For the PS model, the correlated component is most strongly degenerate with the ACC-noise amplitude ( $s_{acc}$ ).
Fisher Matrix Analysis: The direction of the bias aligns with the Fisher matrix parameter-correlation coefficients. The incomplete model preferentially absorbs the missing covariance through the parameter whose spectral effect most closely resembles the omitted contribution.

Significance and Claims
The paper demonstrates that correlations in test-mass ACC noise can constitute a relevant systematic for space-based SGWB searches. The authors claim that:

Neglecting nonzero ACC correlations can produce systematic parameter biases even if the signal and instrumental auto-noise models are otherwise correctly specified.
The direction of the bias is determined by the frequency dependence of the correlation spectrum, not just its amplitude.
Explicit inclusion of correlated noise in mission-realistic noise models and inference pipelines is necessary to avoid biased inference.

The authors maintain a modest scope, noting that their conclusions are based on phenomenological models (WN and PS) rather than first-principles subsystem models. They acknowledge that the extension to $\rho=1$ is diagnostic rather than a realistic in-flight expectation and that the study assumes equal link-level noise and a diagonal AE likelihood. Future work is suggested to incorporate more realistic, frequency-dependent cross spectra and unequal noise levels into the full AET covariance.

More like this