At the stellar noise frontier: a transit survey of 121 TESS M3--M6 dwarfs

This paper presents a systematic transit survey of 121 newly accessible mid-to-late M-dwarfs using TESS Cycle 6+ data, which identified 20 transit-like signals across 16 systems and established a robust reliability framework to prioritize two high-confidence candidates for radial velocity follow-up while flagging others for further observation.

The Gist

Imagine the universe as a giant, noisy ocean. For decades, astronomers have been trying to find "islands" (planets) by watching the water level drop slightly when a boat passes by. But the ocean is often rough, with waves and storms (stellar activity) that make it hard to tell if a dip in the water is a boat or just a random wave.

This paper is about a team of astronomers who decided to look at a specific, very tricky part of the ocean: M-dwarf stars. These are small, cool, red stars. They are the most common stars in our neighborhood, and because they are so small, a planet passing in front of them creates a very deep, noticeable dip in light—like a large boat passing in front of a tiny rowboat.

However, there was a problem. Most of these stars had only been watched for a short time by the TESS space telescope. It's like trying to predict the weather by looking at the sky for only 20 minutes; you might miss a storm or a sunny day that happens later.

The "Newly Enabled" Discovery

With the telescope recently taking more data (specifically "Cycle 6+"), a new group of these red stars finally had enough observation time to be studied properly. The authors call them "Newly Enabled" targets.

Think of it like this: Imagine you are trying to hear a whisper in a noisy room. For years, you only had 10 seconds of silence to listen. Now, thanks to new technology, you have 10 minutes. Suddenly, you can hear things you never could before. This paper is the report on what they heard during those extra minutes.

The Detective Work: Hunting for Planets

The team built a super-smart computer pipeline (a set of rules and algorithms) to act as a detective. Here is how they worked, using simple analogies:

1. The Net (TLS Algorithm): They used a tool called TLS (Transit Least Squares) to cast a wide net over the data, looking for any pattern that looked like a planet passing in front of a star.
2. The Filter (The 18-Check Cascade): Finding a dip isn't enough. The ocean is full of "fake" dips caused by the star's own mood swings (flares, spots, and rotation). The team created an 18-step filter.
   • Analogy: Imagine you hear a knock on the door. Is it a pizza delivery (a planet)? Or is it the wind (noise)? Or your cat scratching (a star spot)? The filter checks: "Did the knock happen at the right time? Is it too loud? Does it sound like a cat?" If it fails any check, it's tossed out.
3. The Noise Test (The Reliability Framework): This is the most unique part of the paper. Since these stars are so "noisy," the team invented a way to test if a signal is real or just a trick of the light. They used three tests:
   • The Shuffle Test: They shuffled the data around like a deck of cards. If the "planet" still appeared after shuffling, it was just noise.
   • The Mirror Test: They flipped the data upside down. If a "bump" appeared where the "dip" should be, it proved the signal was just random noise.
   • The Scramble Test: They scrambled the rhythm of the data. If the pattern survived, it was likely real.

The Results: What Did They Find?

Out of 121 stars they studied, they found 20 potential planet signals. None of these had been found before!

However, because the stars are so noisy, they had to be very careful about what they claimed to be a "planet." They sorted the candidates into three tiers, like a video game difficulty setting:

• Tier 1 (The "Gold" Candidates): These are the most promising. The signal was strong, passed all the noise tests, and looks very much like a real planet. There are 2 of these. They are the top priority for future telescopes to confirm.
• Tier 2 (The "Silver" Candidates): These look promising but need a little more data to be sure. There are 7 of these.
• Tier 3 (The "Noise" Candidates): These signals are so weak that they might just be the star acting up. They are currently indistinguishable from the background noise. There are 10 of these. They need more observation time to prove they are real.

The Big Picture: Why This Matters

This paper is important for a few reasons:

1. It maps the "Noise Frontier": The authors admit that for these specific types of stars, it is incredibly hard to find small planets because the stars are so active. They quantified exactly how hard it is. They found that about 17% of the signals they found were actually just noise (false alarms).
2. It sets a new standard: They created a new way to test if a signal is real, which other astronomers can use for other difficult stars.
3. It points the way forward: The paper concludes that to be absolutely sure about the "Tier 3" candidates, we need to keep watching these stars for longer. If the TESS telescope keeps observing them, the "signal" will get louder, and the "noise" will become clearer.

In a Nutshell

This paper is a report from the front lines of astronomy. The team went into a very noisy, difficult environment (active red dwarf stars) with a new set of tools (more telescope time and better filters). They found some very promising "whispers" of new planets, but they are honest enough to say, "We aren't 100% sure yet because the room is so loud." They have identified the best whispers to listen for next, and they have taught us how to tell the difference between a planet and a star's mood swings.

Technical Summary

Here is a detailed technical summary of the paper "At the stellar noise frontier: a transit survey of 121 TESS M3–M6 dwarfs" by Yohann Tschudi.

1. Problem Statement

M-dwarf stars are ideal hosts for detecting small, Earth-sized transiting planets due to their small radii. However, mid-to-late M-dwarfs (spectral types M3–M6, $T_{eff} \approx 2700–3400$ K) present a significant detection challenge when observed with the Transiting Exoplanet Survey Satellite (TESS) under specific conditions:

• Sparse Temporal Coverage: Many targets only recently accumulated sufficient multi-sector coverage (via TESS Cycle 6+, Sectors 80+) to allow for the detection of planets with longer periods ($P = 10–100$ days). Previously, these stars had $\le 2$ archival sectors, limiting searches to short periods ($P \lesssim 7$ days).
• Stellar Noise Frontier: Active M-dwarfs exhibit high levels of intrinsic variability (spots, flares, BY Dra modulation) that create a "noise floor." This noise often mimics transit signals, leading to high false alarm rates (FAR) that standard detection thresholds (e.g., Signal Detection Efficiency, SDE $\ge 7$) cannot reliably distinguish from genuine planetary transits.
• Gap in Knowledge: Existing pipelines are optimized for bright targets or short periods. There is a lack of systematic characterization regarding the detection limits and false alarm rates for small planets around active, faint M-dwarfs with sparse, multi-sector baselines.

2. Methodology

The study employs a rigorous, multi-stage pipeline designed specifically for the "noise frontier" of active M-dwarfs.

A. Target Selection (The "Newly Enabled" Sample)

A 9-step selection funnel was applied to 498,312 TIC M-dwarfs to isolate 121 targets that were previously unsearchable at extended periods but became accessible with recent TESS data:

• Spectral Type: Restricted to M3–M6 ($T_{eff} = 2700–3400$ K) to ensure deep transit depths ($>800$ ppm for Earth-sized planets).
• Coverage Criteria: Targets must have $\le 2$ archival sectors but $\ge 1$ Cycle 6 sector, resulting in a baseline ($B$) of $>100$ days (up to 2,600 days) despite having only $\sim 80–160$ days of actual photometry.
• Quality Filters: Excluded bright stars (saturation/red noise), faint stars (photon noise), unresolved binaries (Gaia RUWE $< 1.4$), and targets with known planets or high contamination ratios.

B. Detection Pipeline

1. Preprocessing: TESS photometry (SPOC/QLP) is detrended using Gaussian Processes (GP) with a two-component kernel (stochastic harmonic oscillator for variability + Matérn-3/2 for trends) to remove stellar activity while preserving transit shapes.
2. Detection Algorithm: Uses Transit Least Squares (TLS) with a "Scout/Sniper" optimization strategy to handle long baselines efficiently.
   • Scout: Binned lightcurve search with relaxed threshold ($SDE \ge 4.5$).
   • Sniper: Unbinned refinement of Scout candidates.
   • Threshold: Effective SDE ($SDE_{eff}$) $\ge 7.0$, penalized for red noise ($\beta_{rn}$).
3. Harmonic Correction: A three-mode evidence system (equal-depth secondary, odd-even depth mismatch, equal-depth thirds) resolves period aliases (e.g., $2P$,$3P$) common in active stars.

C. Validation Cascade (18-Check System)

Before statistical vetting, candidates undergo 18 diagnostic checks to reject false positives:

• Physical/Instrumental: Signal polarity, period aliases (Earth rotation, spacecraft orbit), centroid offsets, and secondary eclipses.
• Stellar Variability: Tests for sinusoidal variability (SWEET), out-of-transit scatter, and detrending response to ensure signals are not artifacts of stellar rotation or flares.
• System Parsimony: Caps the number of plausible planets per star based on stellar activity levels (MAD).

D. Candidate Assessment

1. Statistical Vetting: TRICERATOPS calculates False Positive Probabilities (FPP) and Nearby False Positive Probabilities (NFPP) to rule out background eclipsing binaries (BEBs).
2. Physical Verification: Gaia DR3 astrometry (RUWE and contaminant search) directly tests for unresolved binaries within $30''$.
3. Signal Reliability Framework (Novel Contribution): Three independent empirical tests quantify the robustness of the signal against the host star's specific noise floor:
   • Circular Shift FAR: Shuffles sectors to destroy transit coherence while preserving stellar variability. If a candidate appears in the scrambled data, it is a false alarm.
   • Lightcurve Inversion: Reflects the flux ($f_{inv} = 2 - f$). If a signal appears in the inverted data, the noise floor mimics a transit.
   • Fourier Phase Scrambling: Randomizes FFT phases to preserve the power spectrum but destroy waveform coherence, estimating the False Alarm Probability (FAP).

3. Key Contributions

• The "Newly Enabled" Survey Strategy: First systematic survey targeting M3–M6 dwarfs that only recently crossed the multi-sector detection threshold, exploring the $P=10–100$ day regime for these specific stars.
• 18-Check Validation Cascade: A comprehensive diagnostic suite tailored to M-dwarf specific false positives (BY Dra variability, harmonic nests).
• Three-Test Reliability Framework: A novel classification system (Tier 1–3) that moves beyond binary "detected/not detected" to quantify the probability that a signal is distinguishable from the host star's intrinsic noise.
• Empirical Characterization of the Noise Frontier: Quantifies the false alarm rate and detection completeness limits for active M-dwarfs with sparse coverage.

4. Results

• Pipeline Validation: Tested on 10 known systems (16 planets), achieving 100% recovery with zero false positives.
• Survey Statistics:
  • Analyzed 121 stars.
  • Detected 20 transit-like signals across 16 systems (none had prior TOI designations).
  • Rejection Rate: 39% of initial candidates (13/33) were rejected, primarily due to eclipsing binary scenarios.
• Signal Reliability Classification:
  • Tier 1 (High Robustness): 2 candidates (both from TIC 211401412). These have clean FAR tests, inversion tests below threshold, and 0% Fourier FAP. These are the highest priority for Radial Velocity (RV) confirmation.
  • Tier 2 (Moderate Robustness): 7 candidates. Require additional photometric monitoring or pre-screening.
  • Tier 3 (Noise-Susceptible): 10 candidates. Signals sit at or above the noise floor; require additional TESS sectors to confirm persistence.
  • Monotransit: 1 candidate (TIC 72750190) excluded from the tier system due to single-epoch detection.
• False Alarm Rate (FAR): The global empirical FAR for the survey is 17.4% (21/121 stars produced at least one spurious signal).
• Injection-Recovery Tests:
  • Calm Stars: 56% overall recovery; 85% for $R_p \ge 2.0 R_\oplus$.
  • Active Stars: Recovery drops to 18%; no planets $< 1.76 R_\oplus$ recovered.
  • Period Dependence: Recovery for $P \ge 20$ days drops to 0% for active stars and 29% for calm stars due to sparse coverage ($\le 3$ transits).

5. Significance and Conclusions

• Defining the Limits: The study confirms that for active M3–M6 dwarfs with sparse TESS coverage, no candidate signal significantly exceeds the host star's noise floor. The detection of small planets is currently limited by stellar variability rather than photon noise.
• Resource Allocation: The Tier system provides a clear roadmap for follow-up:
  • Tier 1: Immediate RV follow-up.
  • Tier 2: Continued TESS monitoring or RV pre-screening.
  • Tier 3: Requires more TESS data (accumulating sectors) to establish signal persistence before investing in expensive follow-up.
• Scalability: The methodology (validation cascade and reliability framework) is directly transferable to:
  • Earlier M-dwarfs (K5–M2) entering multi-sector coverage.
  • Young, pre-main-sequence stars with extreme variability.
  • Ultracool dwarfs (M7–L) observed by ground-based surveys.
  • The faint end of the PLATO mission sample.
• Future Outlook: The TESS Continuous Viewing Zone (CVZ), with 30–44 sectors, is predicted to provide a $\sim 3.4\times$ gain in SDE, potentially resolving the ambiguity for the Tier 2 and 3 candidates identified in this survey.

In summary, this paper establishes a robust framework for navigating the "stellar noise frontier," demonstrating that while active M-dwarfs are challenging, a combination of rigorous vetting and empirical reliability testing can isolate the most promising candidates for small planet discovery.