TL;DR
This paper introduces a simulation to investigate the limitations of empirical templates for precise radial velocity (RV) measurements of M dwarf stars. They discover that imperfect interpolation during template creation introduces a fundamental noise floor, limiting RV precision to ~0.5-0.8 m/s for instruments like SPIRou, even in the absence of photon noise. This highlights a critical bottleneck for detecting low-mass exoplanets around these stars.
What Problem Does This Paper Address?
Detecting exoplanets, especially Earth-like ones, relies heavily on the radial velocity (RV) technique, which measures tiny shifts in a star's light caused by a planet's gravitational tug. M dwarf stars are prime targets because their smaller size means even small planets induce larger RV signals. However, achieving the necessary precision (~1 m/s or better) is incredibly challenging. A key method for this is using "empirical templates" – essentially, a highly averaged spectrum of the star that represents its intrinsic, non-shifted light. This template is then compared to individual observations to measure RV shifts. The problem is that M dwarfs have complex, blended spectral lines in the near-infrared (NIR), making accurate spectral modeling difficult. This paper investigates whether these empirical templates, while seemingly robust, introduce their own subtle errors that limit RV precision, particularly for M dwarfs.
Methods
The authors developed a numerical simulation to mimic RV observations from the SPIRou instrument. This simulation starts with high-resolution PHOENIX stellar spectra, which are then Doppler-shifted to simulate RV variations (including a hypothetical planet). Crucially, these spectra are then convolved to the instrument's spectral resolution and, importantly, interpolated to match the detector's pixel grid. Photon noise is added to create realistic synthetic observations.
The core of their investigation lies in how these synthetic observations are used to build empirical templates. They take a subset of these observations, align them (after correcting for barycentric velocity), and then take the median to form a template. This template is then used to estimate RVs from the remaining synthetic observations. By comparing the RVs derived from the template to the known injected RVs (and also to RVs derived directly from the intrinsic spectrum), they can quantify the impact of the template creation process, specifically focusing on the effects of interpolation.
Datasets
The study utilizes PHOENIX stellar spectra as the basis for generating synthetic M dwarf spectra. These spectra are chosen for their high resolution and comprehensive modeling of stellar atmospheres. No new observational datasets were used; instead, the research relies entirely on a simulated environment to isolate and study specific error sources.
Key Insights and Ideas
- Interpolation is a hidden noise source: The most significant finding is that the process of interpolating spectral data to match detector pixel grids, even in the absence of photon noise, introduces subtle distortions in spectral line shapes. These distortions are carried over into empirical templates.
- Templates aren't perfect proxies: Empirical templates, while useful, do not perfectly represent the true intrinsic stellar spectrum. They contain "interpolation residuals" that can mimic stellar jitter or activity, leading to biased RV measurements.
- M dwarfs are particularly susceptible: Cooler M dwarfs, with their denser and more complex spectral features, are more prone to larger interpolation errors. This means their inherent RV advantage (larger signals for small planets) can be partially negated by template-based RV analysis limitations.
- Sampling resolution is critical: The degree of interpolation error is directly tied to the detector's sampling resolution. Instruments with lower sampling (fewer pixels per spectral feature) suffer more from these errors.
- A fundamental RV floor: For instruments like SPIRou, the combination of spectral complexity and detector sampling imposes an intrinsic RV uncertainty floor of approximately 0.5-0.8 m/s, independent of stellar brightness or telescope aperture. This is a significant limitation for detecting sub-m/s RV signals from Earth-like planets.
Key Results
- The simulation demonstrates that interpolation errors can dominate over photon noise at high signal-to-noise ratios (S/N).
- For a typical M4V star observed with SPIRou, the interpolation error imposes an RV uncertainty floor of 0.5-0.8 m/s.
- The magnitude of the interpolation error increases for cooler M dwarfs (e.g., M9V) compared to hotter ones (e.g., M5V) at the same sampling resolution.
- Higher detector sampling resolution (more pixels per spectral feature) significantly reduces interpolation errors, with instruments like NIRPS achieving a much lower floor (~0.07 m/s).
Limitations and Future Work
The authors acknowledge that their simulation simplifies reality by neglecting other noise sources like telluric contamination and stellar activity, which can further compound errors. They also assume a Gaussian point spread function (PSF) for the instrument, and real spectrographs may have PSFs that are less amenable to cubic spline interpolation, potentially leading to underestimated interpolation errors.
Future work should focus on developing improved spectral modeling techniques that can better capture the nuances of stellar spectra without relying on interpolation. The authors suggest exploring forward-modeling approaches, where stellar spectra are expressed as a sum of basis functions, rather than relying on template matching. Additionally, they emphasize the need for detector sampling beyond the Nyquist rate to mitigate these interpolation-induced errors and achieve the next level of RV precision required for detecting Earth-like exoplanets.
Summary updated