A Gaussian process model for stellar activity in 2-D line profile time-series

Haochuan Yu, S. Aigrain, Baptiste Klein, M. Cretignier, Florian Lienhard, Stephen Roberts

TL;DR

Researchers have developed a new physics-driven Gaussian Process (GP) framework that models stellar activity directly within 2D line profile time-series. By exploiting the correlation between time and velocity (wavelength) simultaneously, this method can successfully disentangle tiny planetary signals—as small as 0.3 m/s—from the "noise" of stellar spots and faculae, even when the planet's orbit perfectly matches the star's rotation period.

What Problem Does This Paper Address?

The search for Earth-like exoplanets primarily relies on the Radial Velocity (RV) method, which detects the tiny "wobble" a planet induces in its host star. However, stars are not static light bulbs; they are dynamic, boiling spheres of plasma. Features like sunspots (cool, dark regions) and faculae (bright, magnetic regions) distort the shapes of spectral lines.

These distortions create "activity signals" that can be orders of magnitude larger than the signal from a small, rocky planet. To make matters worse, these signals often repeat on timescales similar to planetary orbits, leading to "false positives" or completely masking real planets. Current mitigation strategies usually look at either the time domain (how the total velocity changes over months) or the velocity domain (how the line shape changes at a single moment). This paper addresses the need for a hybrid approach that uses all the information available in the 2D "data cube" of spectral line profiles over time.

Methods

The authors introduce the CCF-GP framework. Instead of reducing a complex spectral line to a single number (like its center or width), they model the entire Cross-Correlation Function (CCF) as it evolves.

Key components include:

  • Physics-Driven Modeling: The model is an extension of the $FF'$ framework. It treats the line profile perturbation as a sum of two physical effects: the photometric effect (flux blocked by a spot) and the Inhibition of Convective Blueshift (ICB) (magnetic fields slowing down the upward flow of plasma).
  • Taylor Expansion: They approximate the distorted line profile using a Taylor series expansion of the "quiet" stellar profile. This allows them to represent complex shape changes as a combination of the profile's derivatives.
  • Latent Gaussian Processes: The time-varying coefficients of these derivatives are modeled using a latent GP and its first derivative. This captures the quasi-periodic nature of stellar rotation and active region evolution.
  • Joint Inference: The model jointly samples planetary orbital parameters and stellar activity parameters using PolyChord, a nested sampling algorithm. This ensures that the activity model doesn't "eat" the planetary signal during the cleaning process.

Datasets

The framework was validated using three distinct types of data:

  1. SOAP 2.0 Simulations: Synthetic CCFs generated with known spots, faculae, and injected planets. This provided a "ground truth" to test if the model could break the degeneracy between activity and planets.
  2. HARPS-N Solar Data: Two years of extremely high-quality observations of our own Sun, treated as a distant star. The authors performed "injection/recovery" tests by adding simulated planetary signals to this real-world noise.
  3. HD 13808 Observations: Archival data from a real K-type star known to host two Neptune-mass planets, used to demonstrate efficacy on standard observatory data with a Signal-to-Noise Ratio (SNR) of ~100.

Key Insights and Ideas

  • Breaking the Period Degeneracy: One of the most significant "aha" moments was demonstrating that the model can distinguish a planet from stellar activity even when their periods are identical. Because a planet causes a pure shift while activity causes a shape distortion, the 2D model sees the difference that 1D methods miss.
  • Information Retention: By working at the CCF level rather than the RV level, the model retains "shape" information that is usually discarded. This extra dimensionality acts as a powerful constraint on the GP, preventing it from being too flexible and over-fitting the data.
  • The Power of Derivatives: Using the derivatives of the line profile (from the Taylor expansion) provides a physically motivated basis set. It turns out that most activity-induced distortions can be remarkably well-approximated by just the first few derivatives of the quiet-sun profile.

Key Results

  • Sub-Meter/Second Precision: In the solar data tests, the method accurately recovered the signal of a 1.5-Earth mass planet with a semi-amplitude of only 0.3 m/s and a 33-day period, even during high solar activity.
  • Superiority over Multi-GP: In head-to-head tests against standard "multi-GP" models (which use 1D indicators like FWHM or Bisector Span), the CCF-GP provided more accurate estimates of planetary phase and amplitude.
  • Robustness at Moderate SNR: The method remained effective at SNRs down to ~100, making it applicable to a wide range of existing and future spectroscopic surveys, not just the "ultra-quiet" brightest stars.

Limitations and Future Work

While powerful, the authors note several areas for improvement:

  • Gaussianity Assumption: The current model assumes the noise in the CCF time-series is Gaussian. In reality, stellar processes like super-granulation might introduce non-Gaussian "flicker" that the model doesn't yet account for.
  • Simplified Physics: The model treats the ratio of spot-to-plage area as constant over time, which isn't strictly true as active regions evolve.
  • Computational Intensity: Using nested sampling on 2D data is computationally expensive. Future iterations may need to explore faster GP kernels or more efficient sampling techniques to scale up to larger datasets.
  • Differential Line Response: Different spectral lines respond to activity in different ways (e.g., some are more sensitive to magnetic fields). Incorporating this "line-by-line" physics into the 2D framework is a logical next step.

Summary updated

Abstract

Stellar active regions like spots and faculae can distort the shapes of spectral lines, inducing variations in the radial velocities that are often orders of magnitude larger than the signals from Earth-like planets. Efforts to mitigate these activity signals have hitherto focused on either the time or the velocity (wavelength) domains. We present a physics-driven Gaussian process (GP) framework to model activity signals directly in time series of line profiles or Cross-Correlation Functions (CCFs). Unlike existing methods which correct activity signals in line profile time series, our approach exploits the time correlation between velocity (wavelength) bins in the line profile variations, and is based on a simplified but physically motivated model for the origin of these variations. When tested on both synthetic and real data sets with signal-to-noise ratios down to $\sim$ 100, our method was able to separate the planetary signal from the activity signal, even when their periods were identical. We also conducted injection/recovery tests using two years of realistically sampled HARPS-N solar data, demonstrating the ability of the method to accurately recover a signal induced by a 1.5-Earth mass planet with a semi-amplitude of 0.3 m/s and a period of 33 days during high solar activity.
BibTeX
@inproceedings{yu2024gaussian,
  author = {Haochuan Yu and S. Aigrain and Baptiste Klein and M. Cretignier and Florian Lienhard and Stephen Roberts},
  title = {{A Gaussian process model for stellar activity in 2-D line profile time-series}},
  booktitle = {Monthly},
  year = {2024},
  doi = {10.1093/mnras/stae2421},
}
Figure