No Helium Detected in LHS 1140 b from Four JWST NIRISS/SOSS Transits

This paper presents four JWST NIRISS/SOSS transit observations of the exoplanet LHS 1140 b, obtained between 2023 and 2026, to search for the metastable helium triplet at 1.0833 µm. The study was motivated by a ground-based detection of helium absorption reported by C. Cherubim et al. (2026) using the WINERED spectrograph on the Clay/Magellan II telescope, which suggested the planet may have a helium-dominated upper atmosphere undergoing escape. However, the authors detect no helium absorption in any of the four JWST visits.

The paper states: "We detect no helium absorption in any visit. We reject the best-fit ground-based model at > 3σ in each visit, and find no clear trend in mass-loss with time. Our results suggest the reported ground-based detection may be spurious, although variability cannot be excluded if detectable helium absorption occurs in ≲ 50% of transits."

The four transits include two archival observations from December 2023 (program DD 6543) and two new observations from August 2025 and July 2026 (program GO 7073). All observations used the SUBSTRIP256 subarray with NISRAPID readout. The data were reduced with the FIREFLy pipeline, and the two new transits were independently reduced with the exoTEDRF pipeline to confirm results.

The analysis focused on the narrow wavelength range of 1.07–1.095 µm around the helium triplet. The authors produced column-level light curves to maximize signal-to-noise. They found no excess absorption in the helium light curves compared to white light curves in any visit, nor any evidence for pre- or post-transit helium tails. The residuals between helium and white light curves were structureless.

The authors calculated 2σ upper limits on helium absorption depth at both SOSS resolution and the ground-based WINERED resolution. These upper limits ranged from 0.02% to 0.04% at SOSS resolution and 0.41% to 0.61% at WINERED resolution across the four visits. These limits are similar to the 0.6% upper limit reported in the 2025 ground-based nondetection.

To compare with the ground-based detection, the authors convolved the best-fit pwinds model from C. Cherubim et al. (2026) to SOSS resolution. They found that if the reported 1.24% absorption depth were persistent, it would be confidently detectable in each SOSS visit. However, no helium absorption was seen. The χ2 analysis showed each SOSS transit rejects the ground-based model at > 3σ, and combining all four transits rejects it at 9.9σ.

The authors also used pwinds to place upper limits on possible mass-loss rates, varying only the mass-loss rate while keeping all other parameters fixed to those from C. Cherubim et al. (2026). Their 2σ upper limits ranged from 5.4 × 106 to 3.1 × 107 g s−1 across the four visits. Combined with the ground-based results, the single reported detection of 2 × 108 g s−1 from 2024 stands out as an outlier, with all nondetections reporting consistent upper limits on the order of 106–107 g s−1. The authors note there is no clear trend in mass-loss rate over time across all six observations.

The paper discusses the possibility of time-variable escape. If the helium signal occurred during only a random fraction of time f, then for a duty-cycle of f = 50%, four nondetections with SOSS would be expected only 6.3% of the time, implying f < 53% at 2σ. The authors conclude that time-variability cannot be readily rejected, but the reported 2024 signal is unlikely to be persistent.

The authors note several caveats: at very high mass-loss rates (>109 g s−1), the helium feature decreases due to hydrogen self-shielding, so their nondetections are technically consistent with such extreme rates, though unlikely given the system's age (>5 Gyr). They also acknowledge degeneracies in the pwinds modeling parameters and that factors like stellar EUV flux and upper-atmosphere temperature are model-dependent.

The paper concludes that the helium nondetections cast doubt on the helium-dominated upper atmosphere interpretation of LHS 1140 b, but do not constrain whether the planet is a mini-Neptune or a water world. One more SOSS transit is scheduled for November 2026. The authors state: "with six currently available observations, it is now incontrovertible that the previously detected helium absorption is not persistent, and if that signal is time-varying, then it recurs less than half of the time. At this point, it is unlikely though still possible that the ground-based signal represented a true escaping atmosphere from the ever-mysterious LHS 1140 b."

Improvements to AI Systems:

1. Astronomical Data Reduction & Cross-Validation Pipeline

• Implement an AI that automatically reduces JWST/NIRISS time-series data using multiple independent pipelines (e.g., FIREFLy and exoTEDRF) and cross-checks results for consistency, flagging discrepancies that could indicate systematic errors.

• The improved system can autonomously produce column-level light curves, perform white-light vs. spectral-light curve differential analysis, and generate residual maps to detect structureless noise vs. real signals.

2. Automated Exoplanet Atmospheric Escape Detection & Model Rejection

• Build an AI that ingests ground-based and space-based transit spectra, fits atmospheric escape models (e.g., pwinds), and computes chi-squared rejection statistics for competing hypotheses (e.g., persistent vs. time-variable helium absorption).

• The system can automatically calculate upper limits on absorption depth at multiple spectral resolutions and convert nondetections into mass-loss rate constraints, including handling degeneracies and self-shielding effects.

3. Time-Variability & Duty-Cycle Inference

• Develop a Bayesian or frequentist AI module that, given a set of detections and nondetections across multiple epochs, estimates the probability that a signal is persistent vs. sporadic, including computing duty-cycle upper limits (e.g., f < 53% at 2σ) and predicting the likelihood of future detections.

• The improved system can schedule optimal future observations (e.g., deciding when to use the November 2026 transit) to maximize the chance of confirming or refuting time-variable escape.

4. Multi-Wavelength Stellar Contamination & False-Positive Screening

• Create an AI that automatically screens for stellar activity, telluric contamination, and instrumental systematics by comparing helium line light curves to adjacent continuum bands and white-light curves, rejecting spurious ground-based detections with quantified confidence (e.g., >3σ per visit, 9.9σ combined).

• The system can flag observations where the signal-to-noise is insufficient to detect a predicted absorption depth, and recommend binning or resolution adjustments (e.g., SOSS vs. WINERED) to optimize detectability.

5. Physical Parameter Inference with Degeneracy Handling

• Implement an AI that, given nondetections, explores the full parameter space of escape models (mass-loss rate, EUV flux, upper-atmosphere temperature, helium abundance) using nested sampling or MCMC, and outputs marginalized upper limits while explicitly reporting degeneracies and model-dependent caveats (e.g., high mass-loss >109 g s−1 where helium weakens).

• The improved system can distinguish between no escape and extreme escape with self-shielding scenarios, and provide probabilistic statements about the planet's nature (mini-Neptune vs. water world) without overinterpreting null results.

6. Automated Literature Integration & Outlier Detection

• Build an AI that ingests all published detections and nondetections for a target (e.g., LHS 1140 b), normalizes them to common metrics (absorption depth, mass-loss rate), and performs outlier analysis to identify inconsistent claims (e.g., the 2024 ground-based detection vs. four JWST nondetections).

• The system can generate a consensus view, quantify the statistical tension between datasets, and produce a real-time updated confidence map for the scientific community.

7. Observation Planning & Prediction

• Use the AI to simulate synthetic JWST/SOSS observations under various escape scenarios (persistent, time-variable with duty cycle f, extreme self-shielding) and predict the detectability of helium in upcoming transits, optimizing exposure time, subarray choice, and binning.

• The improved system can recommend whether additional transits are needed to reach a conclusive answer (e.g., to push duty-cycle limits below 10%) and estimate the required number of visits.

What the Improved AI System Can Do:

• Automatically reduce, validate, and analyze JWST exoplanet transit data from raw time-series to final physical constraints, with minimal human intervention.

• Provide statistically robust, reproducible rejections of spurious ground-based detections and quantify upper limits on atmospheric escape rates, including time-variability.

• Guide future observational campaigns by predicting detectability and optimizing telescope time, while transparently reporting model dependencies and degeneracies.

In the effort to determine which low-mass exoplanets have atmospheres, LHS 1140 b remains one of the most favorable targets. Its large size (5.6 M and 1.7 R) and relatively long orbital period (24.7 days) imply an atmosphere may be likely, and notably, recent interior models favor either a hydrogen-dominated "mini-Neptune" or a "water world" over a true terrestrial planet. Another possibility is that it has a helium-rich atmosphere. This hypothesis is supported by recent ground-based observations that detected the metastable helium triplet during transit. These observations indicated there may be current helium escape from the planet's upper atmosphere, yet the signal was not detected during a subsequent observation, suggesting time-variable escape. Here we present four observations of LHS 1140 b with JWST NIRISS/SOSS, which covers the metastable helium triplet, obtained between 2023 and 2026. These observations span the epoch of the ground-based measurements, and although none were contemporaneous with the ground-based transits, all four are sensitive to helium absorption at the previously reported level. However, we detect no helium absorption in any visit. We reject the best-fit ground-based model at >3 sigma in each visit, and find no clear trend in mass-loss with time. Our results suggest the reported ground-based detection may be spurious, although variability cannot be excluded if detectable helium absorption occurs in 50% of transits. The nature of LHS 1140 b thus remains a mystery until future transmission and emission analyses are complete.