Machine Learning Reveals Hidden Subgroups Within Pulsating Stars, Rewriting Stellar Classification

Scuti stars, pulsating variable stars crucial for modelling stellar evolution, are currently categorised by their light curve amplitude. However, J. R. Rodon, J. Pascual-Granado, and M. Lares-Martiz, from the Departament d’Astronomia i Astrofísica, Universitat de Valéncia, along with M. Rodríguez Sánchez and C. Roche et al., demonstrate that this approach may oversimplify the complex non-linear phenomena driving their pulsations. Their research applies machine learning techniques to a sample of 142 Scuti stars observed by space-based missions, analysing frequency-domain features and non-linear combinations to identify intrinsic subgroups. This work is significant because it reveals additional groupings beyond the conventional High-Amplitude and Low-Amplitude classifications, suggesting a richer diversity of internal physical mechanisms and resonance effects than previously understood.

Refining δ Scuti star classification through frequency analysis and machine learning offers improved accuracy and efficiency

Scientists are employing machine learning to reassess the classification of δ Scuti stars, a type of pulsating variable star crucial for understanding stellar evolution and internal structures. Current categorization relies on peak-to-peak amplitude of light curves, dividing these stars into High-Amplitude δ Scuti (HADS) and Low-Amplitude δ Scuti (LADS) based on a 0.3 magnitude threshold.
However, this method may oversimplify the complex interplay of pulsation mechanisms and non-linear effects within these stars, potentially leading to inaccurate classifications. This investigation challenges the amplitude-based system by exploring frequency-domain characteristics and non-linear phenomena to identify intrinsic subgroups among δ Scuti stars.

The research utilizes a sample of 142 δ Scuti stars observed by space telescopes including CoRoT, Kepler, and TESS, applying hierarchical clustering with Ward’s linkage to analyze their light curves. Data processing involved the Best Parent Method, a technique designed to extract non-linearities and identify parent and child frequencies within the stellar pulsations.

By focusing on fundamental and overtone modes, alongside non-linear features like harmonic, sum, and subtraction frequencies, researchers aim to uncover previously hidden groupings within the δ Scuti population. Results demonstrate a partial alignment between the existing amplitude-based classification and the clusters identified through frequency-domain features.

Significantly, the study revealed additional subgroups, suggesting a greater diversity of non-linear effects than previously accounted for by amplitude alone. The number of subtraction combinations, a non-linear feature, appears particularly important, potentially indicating resonance effects or other internal physical mechanisms at play within these stars. This work highlights the need for a more nuanced understanding of δ Scuti pulsations, moving beyond simple amplitude measurements to incorporate the complexities of their internal dynamics.

Pulsation Mode Decomposition and Hierarchical Clustering of Scuti Stars reveal distinct stellar populations

Hierarchical clustering with Ward’s linkage was applied to a sample of 142 Scuti stars to investigate intrinsic subgroups within this stellar population. Light curves obtained from space telescopes including CoRoT, Kepler, and TESS served as the primary data source for this research. These light curves underwent processing and analysis utilising the Best Parent Method, a technique designed to extract non-linearities and identify parent and child frequencies within the data.

This method facilitated the detailed examination of complex pulsation spectra and the identification of subtle variations in stellar luminosity. The methodology centred on the extraction of both frequency-domain features and non-linear features from the processed light curves. Fundamental and overtone modes were identified and quantified, providing information about the primary pulsation frequencies of each star.

In addition to these linear features, harmonic, sum, and subtraction frequencies of the fundamental modes were calculated to characterise non-linear effects. These non-linear frequencies provide insights into interactions between different pulsation modes and between pulsations and the stellar medium.

The resulting set of features, encompassing linear modes and non-linear combinations, was then subjected to hierarchical clustering. Ward’s linkage criterion was employed to minimise the variance within clusters, effectively grouping stars with similar frequency characteristics. This clustering process aimed to reveal underlying relationships between Scuti stars that may not be apparent through traditional amplitude-based classification.

The number of subtraction combinations, a specific non-linear feature, was given particular attention as a potential indicator of resonance effects or other internal physical mechanisms driving stellar pulsations. The work demonstrates that the established amplitude-based classification of HADS and LADS stars only partially aligns with the clusters identified through frequency-domain analysis, suggesting a more nuanced and complex relationship between pulsation amplitude and underlying stellar properties.

Δ Scuti star classifications refined by frequency-domain analysis and non-linear feature assessment reveal subtle pulsational modes

Analysis of 142 δ Scuti stars revealed partial alignment between current amplitude-based classifications and groupings identified through frequency-domain features. The research employed hierarchical clustering with Ward’s linkage to analyze data obtained from the CoRoT, Kepler, and TESS space telescopes.

This work focused on both fundamental and overtone modes, alongside non-linear features such as harmonic, sum, and subtraction frequencies, to discern intrinsic subgroups within the stars. The study identified additional subgroups beyond the traditional High-Amplitude δ Scuti (HADS) and Low-Amplitude δ Scuti (LADS) classifications, suggesting a more complex range of non-linear effects than previously understood.

Specifically, the number of subtraction combinations proved to be a significant feature, potentially indicating resonance effects or other internal stellar mechanisms. Light curves were processed using the Best Parent Method, a technique designed to extract non-linearities and identify parent and child frequencies within the data.

Fundamental and overtone mode frequencies, alongside associated amplitudes and phases, were extracted as key features for the clustering analysis. The BPM algorithm was utilized to identify and remove combination frequencies, ensuring a focused analysis on primary pulsation signals. The clustering results demonstrate that while amplitude remains a relevant factor, it does not fully encapsulate the diversity of pulsation behaviors observed in δ Scuti stars. This suggests that incorporating frequency-domain and non-linear features provides a more nuanced understanding of their internal properties and pulsation mechanisms.

Refining Scuti star classifications through frequency and non-linear pulsation analysis offers improved astrophysical insights

Scuti stars, pulsating variable stars crucial for understanding stellar evolution, have traditionally been classified as either High-Amplitude or Low-Amplitude based on their light curve variations. This established categorization may oversimplify the complex pulsation behaviours and non-linear effects present within these stars, potentially leading to inaccurate classifications.

Recent investigations employed machine learning techniques to analyse a sample of 142 Scuti stars, utilising frequency-domain features and non-linear characteristics to identify intrinsic subgroups and refine current understanding of these stellar objects. The analysis revealed partial alignment between the conventional amplitude-based classification and the clusters identified through frequency-domain features.

However, the inclusion of non-linear features uncovered additional subgroups, indicating a greater diversity of pulsation mechanisms and internal structures than previously recognised. Notably, the number of subtraction combinations, suggestive of resonance effects or other internal physical processes, proved to be a valuable discriminatory factor.

These results demonstrate that relying solely on light curve amplitude can obscure the underlying dynamics of Scuti star oscillations and their associated non-linear effects. This work acknowledges that the traditional amplitude-based classification, while useful, is insufficient to fully capture the complexity of Scuti stars.

Future research will focus on determining the physical origins of the identified clusters by integrating the clustering results with independent observational data and detailed pulsation modelling. This will involve assessing the influence of factors such as stellar rotation, metallicity, binarity, and non-linear mode coupling on the observed pulsation spectra. The application of machine learning to asteroseismology offers a promising avenue for analysing the extensive datasets generated by contemporary and forthcoming space missions, ultimately contributing to more accurate models of stellar interiors and pulsation mechanisms.