Three new exoplanet systems from the Dispersed Matter Planet Project

This paper presents a radial velocity analysis of three bright, low-activity stars identified by the Dispersed Matter Planet Project (DMPP): DMPP-7 (HD 118006), HD 67200, and HD 2134. The authors use a Bayesian framework to compare purely Keplerian models with models incorporating stellar activity via a quasi-periodic Gaussian Process (GP).

• DMPP-7 b: A single 0.72 Saturn-mass giant planet with minimum mass mp sin i = 69.0 ± 1.7 M⊕ and orbital period P = 4.93 d. This is described as a slightly evolved star that harbours a single 0.72 Saturn-mass giant.

• A longer 21 d – 22 d period cannot be conclusively confirmed as a stellar rotation signature rather than a purely Keplerian signal.

• The paper notes: Our Model 2 minimum mass of mp sin i = 69.0 ± 1.7 M⊕ (∆M/M = 2.4%) is 11% higher than the S26 solution, implying DMPP-7 is a 0.72 Saturn mass planet.

• DMPP-7 b lies at the transition between the high-radius population and the Neptunian ridge and savannah regions.

• "HD 67200, which exhibits Ca ii H&K variability, a model with only a GP is strongly favoured over a purely dynamical model."

• The GP model shows moderate evidence for a single Keplerian with P = 2.67 d.

• The paper states: We thus consider the Keplerian signals reported as DMPP-6 b and DMPP-6 c in our previous study (Standing et al. 2026) to be superseded by the models with GPs presented here.

• The tentative planet candidate HD 67200 b has minimum mass mp sin i = 2.07 M⊕ (specifically 2.07+0.52−0.47 M⊕) in a 0.041 AU orbit.

• a 21 d – 32 d rotation period signal is associated with tentative FWHM variability.

• A model with a GP is not conclusively favoured, but all models considered show moderate evidence for an additional single Keplerian with P = 2.78 d.

• The candidate HD 2134 b has minimum mass mp sin i = 2.86 M⊕ (specifically 2.86+0.67−0.65 M⊕) in a 0.0393 AU orbit.

Despite our target selection favouring near edge-on orbital geometries, we find no evidence for transits in TESS photometry. The paper notes: No significant transit signals were found for any of the three targets.

• Observations were made with HARPS at the 3.6m ESO telescope (2015-2023) and ESPRESSO at the VLT (2025).

• The authors used the s-bart code for RV extraction and kima with DNest4 for diffusive nested sampling.

• They compared purely Keplerian models with models including a Gaussian Process using the s+leaf ESP kernel.

• We use a Bayesian framework to compare purely Keplerian models with models incorporating stellar activity via a quasi-periodic Gaussian Process (GP).

The paper concludes: We have found strong Bayesian evidence in support of a 0.72 Saturn mass planet, DMPP-7 b, and moderate evidence for two low mass planet candidates orbiting HD 67200 and HD 2134. The authors emphasize that The evidence for multiplanet systems can be overstated when purely Keplerian models are assumed and that Models that include activity generally yield different configurations with fewer planet candidates, as we have demonstrated for HD 67200.

Further observations are required to establish whether the coherent short-period HD 67200 and HD 2134 signals are stellar or dynamical in origin.

Improvement 1: Bayesian Model Comparison with Activity-Aware Priors

The improved AI system can automatically implement and compare multiple generative models (pure Keplerian vs. Keplerian + quasi-periodic Gaussian Process) using nested sampling, with priors informed by stellar activity indicators (e.g., Ca ii H&K, FWHM, S-index). It will output Bayes factors and posterior probabilities for each model, explicitly flagging cases where activity models supersede planet claims. This prevents overfitting to spurious periodic signals and reduces false-positive exoplanet detections in radial velocity datasets.

Improvement 2: Signal Origin Disambiguation via Multi-Wavelength and Photometric Cross-Check

The AI system can integrate radial velocity (RV) time series with TESS photometry and activity indices to classify each periodic signal as dynamical (planet) or stellar (rotation, spots, convection). For signals like the 21–22 d period in DMPP-7 or the 2.67 d signal in HD 67200, the system will compute the probability that the signal’s phase, amplitude, and coherence match a stellar rotation model vs. a Keplerian orbit, and automatically flag ambiguous cases for follow-up. It can also predict transit windows and geometric alignment probabilities to guide transit searches, as done here with null TESS results.

Improvement 3: Mass–Period–Radius Contextualization for Planet Classification

The AI system can place each detected candidate (e.g., DMPP-7 b at 69 M⊕, P=4.93 d; HD 67200 b at 2.07 M⊕, P=2.67 d; HD 2134 b at 2.86 M⊕, P=2.78 d) into a multidimensional parameter space (e.g., Neptunian ridge, savannah, high-radius population). It will automatically compute the posterior probability of belonging to each population and generate physical interpretations (e.g., photoevaporation, core accretion, tidal migration) based on the host star’s evolutionary state (e.g., slightly evolved for DMPP-7). This aids in prioritizing candidates for atmospheric characterization.

Improvement 4: Adaptive Observation Scheduling for Signal Confirmation

The AI system can design optimal future RV observation campaigns (e.g., HARPS/ESPRESSO) to distinguish between stellar and dynamical origins for ambiguous signals. It will use the posterior distributions from the GP and Keplerian models to compute the expected information gain from each additional measurement, selecting times that maximize phase coverage for the candidate periods (2.67 d, 2.78 d, 21–32 d) while minimizing stellar activity contamination. This reduces the number of observations needed to confirm or refute low-mass planet candidates.

Improvement 5: Automated Reanalysis of Prior Detections with Activity Models

The AI system can re-analyze existing exoplanet catalogs (e.g., from DMPP or other surveys) using the same Bayesian framework with GP activity models. For systems where previous studies reported multiple planets (e.g., DMPP-6 b and c), the system will re-evaluate the evidence and flag cases where a single activity-driven signal was misinterpreted as multiple Keplerians. This improves the reliability of published planet counts and mass estimates, as demonstrated by the supersession of DMPP-6 b/c by a GP-only model.

Improvement 6: Uncertainty-Aware Mass and Orbit Reporting

The AI system will report planet parameters (minimum mass, period, semi-major axis) with full posterior distributions and explicitly separate systematic uncertainties (e.g., activity-induced jitter) from statistical ones. For example, it will output “mp sin i = 69.0 ± 1.7 M⊕ (2.4% uncertainty) under Model 2, but 11% higher than previous S26 solution” and automatically propagate this into habitability or formation models. This prevents overconfident claims and supports reproducible science.

Improved AI System Capability Summary

The enhanced AI system can: (1) run full Bayesian model comparison for RV datasets with stellar activity, (2) classify signal origins using multi-wavelength data, (3) contextualize planet candidates in mass–period–radius diagrams, (4) schedule optimal follow-up observations, (5) re-analyze historical detections to correct false positives, and (6) report parameters with honest uncertainty budgets. This directly reduces false exoplanet claims, improves mass estimates for low-mass planets, and accelerates the confirmation of genuine signals in active star systems.