Discovery and Characterization of Three New High-amplitude delta Scuti Stars from TESS Observations

The following is a detailed summary of the scientific paper, strictly quoting and synthesizing information presented in the text:

Overview and Methodology

The study reports on the discovery and detailed analysis of three new HADS stars, TIC 408074920, TIC 189714989, and TIC 34137913, identified from TESS short-cadence observations. The research utilizes a methodology involving iterative prewhitening analysis on detrended PDC-SAP light curves. The search was conducted in the range of 0–80 c/d, with a significance criterion set at S/N > 5.2, which corresponds to the 4 sigma false-alarm probability level.

Characterization of Individual Stars

• TIC 408074920: This star was observed in TESS Sector 5 over a continuous baseline of 27.4 days. Its light curve exhibits a strongly non-sinusoidal morphology with a peak-to-peak amplitude of approximately 0.3 mag. The dominant mode, f 1 = 14.509721 c/d, is identified as the first overtone radial mode (F1). This identification is supported by the calculated pulsation constant Q, which yields Q = 0.026 plus or minus 0.005 d, consistent with first overtone modes. The star is classified as a HADS variable based on its dominance of a single radial mode and its harmonic series up to 5f 1.

• TIC 189714989: Observed in Sector 17 over 27 days, this target shows large-amplitude (about0.32 mag) variability with distinctly non-sinusoidal morphology. The dominant frequency is f 1 = 9.35128(2) c/d, identified as the fundamental radial mode (F0), supported by Q = 0.034 plus or minus 0.008 d. It possesses a complete sequence of harmonics... up to 8F 0. Crucially, the analysis revealed a series of nearly symmetric side peaks around the harmonic frequencies, of the form n f 1 plus or minus f m, with f m about 0.1197 c/d, which are indicative of amplitude and/or phase modulation. Furthermore, an independent mode, f 24 = 14.1305(2) c/d, was detected; its period ratio P 24/P 0 about 0.662 is inconsistent with a simple radial overtone interpretation and suggests a weakly excited low-degree non-radial p-mode.

• TIC 34137913: Observed in Sector 6, this star's light curve shows high quality. The dominant mode f 1 = 13.131429 c/d is identified as the fundamental radial mode (F0), supported by Q = 0.042 plus or minus 0.006 d, with harmonics detected up to 6F 0. An additional independent mode, f 7 = 17.6174 c/d, was found to have a period ratio P 7/P 0 about 0.746. This ratio is slightly below the canonical range for the first radial overtone, suggesting it may correspond to a low-degree non-radial p-mode or a slightly higher-order radial mode.

Analysis of Variability and Physical Drivers

The study employed a sliding Fourier transform (SFT) analysis using an 8-day window. For all three targets, the amplitudes and phases of the dominant modes remain largely stable within the observational uncertainties over the TESS observing baseline, providing upper limits on short-term variability. However, specific behaviors were noted:

• TIC 408074920: The amplitude of its first overtone mode shows a gradual decrease of approximately about4 mmag over the about27-day TESS baseline. This is interpreted as potentially being related to nonlinear energy exchange among pulsation modes and their harmonics.

• TIC 189714989: The detection of symmetric sidebands, which are a characteristic signature of amplitude and/or phase modulation, suggests a modulation behavior reminiscent of the Blazhko effect commonly observed in RR Lyrae stars.

Stellar Parameter Derivations (SED Fitting and Modeling)

The stellar properties were constrained using two methods:

1. Broadband SED Fitting: This analysis, utilizing ASTROARIADNE2 and incorporating Gaia DR3 parallaxes, provided estimates for T eff, g, [Fe/H], stellar radius (R), and extinction (A V).

2. Stellar Modeling: Evolutionary and seismic models were generated using MESA (version 24.08.1) and the pulsation code GYRE, focusing on radial oscillation modes (=0).

Consistency Checks and Discussion

• SED vs. Seismic Consistency: For TIC 189714989 and TIC 34137913, the stellar parameters derived from the seismic models are broadly consistent with the SED-based results.

• Discrepancy: In contrast, TIC 408074920 shows a significant offset between the two determinations in the HR diagram, specifically a lower luminosity and smaller radius in the seismic model compared to its SED-derived value. This is attributed to the different sensitivities of the two methods.

• Placement: All three stars were placed within the classical instability strip, confirming their classification as delta Scuti variables.

• Conclusion on Complexity: The results confirm that while the pulsation spectra are dominated by radial modes, the data also exhibit additional phenomena that extend beyond the classical picture of HADS stars, including amplitude modulation and non-radial modes. These findings highlight the complex dynamical behaviour of these stars and emphasize the importance of long-term monitoring for future missions like PLATO.

(Note: Given the high stakes, I have structured these improvements into distinct, specialized AI modules. Each module addresses a specific scientific bottleneck derived from the nature of the cited astronomical data—time series variability, multi-source catalog integration, and physical constraint adherence.)

---

Targeting: Variable star classification, period prediction (Paxton et al. 2013, Poretti et al., Ricker et al. TESS data).

Improvement: Development of specialized Convolutional LSTM (ConvLSTM) architectures integrated with Bayesian inference frameworks. Instead of treating the light curve merely as a sequence of numbers, the AI is constrained by fundamental stellar physics equations (e.g., pulsation period-luminosity relations, stellar energy balance).

How it Works: The ConvLSTM processes raw time-series flux data (F(t)) and generates a latent space representation (Z). This Z is then passed through a differentiable physics module (the physics prior) which penalizes any prediction that violates known stellar evolution tracks or pulsation models.

Improved AI Capability:

1. Predictive State Estimation: Accurately predict the future flux state of variable stars and estimate the physical parameters (e.g., effective temperature, luminosity class) that govern that state, even when data gaps or systematic noise are present.

2. Novel Variable Detection: Identify complex variability patterns (e.g., beating modes, multi-periodic behavior) that fall outside pre-defined classification boundaries by analyzing deviations from the established physical priors.

Targeting: Integrating disparate datasets—light curves, spectral data, and large stellar catalogs (Skrutskie et al. 2006, Wright et al. 2010, Ziaali et al. 2019).

Improvement: Implementation of Graph Neural Networks (GNNs) combined with Self-Attention mechanisms to model the relational structure of astronomical objects. The stars are treated as nodes in a graph, and the physical relationships (e.g., co-location, association with a specific stellar population, shared orbital plane) are treated as edges.

How it Works: The MCHCFE ingests multiple feature vectors for each object (e.g., magnitude from Catalog A, pulsation period from Light Curve B, metallicity from Spectrum C). The GNN learns the optimal weighting and fusion method for these features by optimizing the global structure of the galaxy/cluster rather than analyzing objects in isolation.

Improved AI Capability:

1. Robust Membership Determination: Determine the most probable physical association (e.g., Is this variable star truly part of this globular cluster, or is it a foreground interloper?) by analyzing its neighborhood and relationship to known stellar populations, dramatically reducing contamination rates.

2. Feature Imputation: Infer missing or corrupted features for an object (e.g., estimating metallicity [Fe/H]) by leveraging the highly correlated features of its physically connected neighbors within the graph structure.

Targeting: Mitigating instrumental, atmospheric, and survey-specific systematic biases (Twicken et al. 2010, Virtanen et al. 2020 software reliability).

Improvement: Development of a dedicated meta-learning layer that operates on the raw data pipeline outputs. This module uses Transfer Learning principles to model the instrument's response function as a function of time, detector temperature, and observing conditions, rather than just modeling the astrophysical signal.

How it Works: The system is trained on simulated data sets that systematically introduce known instrumental artifacts (e.g., cosmic rays, flat-fielding errors, CCD non-linearity). It learns to predict the residual error map (E sys) for any given observation epoch and subtracts this predicted systematic noise from the raw signal before it reaches the primary analysis module.

Improved AI Capability:

1. High-Fidelity Signal Extraction: Extract astrophysical signals (e.g., faint variability) with unprecedented Signal-to-Noise Ratio (SNR) by actively modeling and compensating for detector non-linearities and temporal systematics, making data from older or less stable surveys usable for modern analysis.

2. Automated Calibration Flagging: Automatically assign a confidence score to every piece of derived data (magnitude, period, etc.) based on the predicted systematic error magnitude for that specific observation point, allowing researchers to filter results with rigorous statistical backing.