Between Plateaus and Slopes: A Data-Driven Exploration of Spectral Diversity Across Type IIP/L Supernovae

TL;DR

The paper addresses whether Type IIP and IIL supernovae form a continuous spectroscopic population or represent distinct classes. It develops a data-driven pipeline using continuum-normalized spectral time series, Gaussian Process-based epoch standardization, and PCA, followed by nonlinear embedding to reveal structure. The main findings show a largely continuous IIP/IIL spectroscopic distribution with a secondary subgroup likely caused by enhanced CSM interaction; spectral diversity also decreases over time, and correlations between light-curve decline rate $s_2$ and spectral features (e.g., H$ ext{alpha}$, Ca II NIR) are robust, aligning with radiative-transfer model expectations that both envelope mass and CSM shape diversity. This approach enables refined classification and potential standardization improvements for SNe II, especially for identifying CSM-influenced objects and constructing spectroscopic twins across epochs.

Abstract

Type II supernovae (SNe II) have been traditionally separated into several subgroups based on their photometric and spectroscopic properties, but whether these represent distinct progenitors or a continuous distribution remains debated. Over the past decade, growing observational evidence has suggested a possible continuity between slow- (IIP) and fast-declining (IIL) SNe. We investigate the continuity of the SNe IIP/L subclasses through a data-driven statistical analysis of spectral time series, aiming to determine whether significant correlations exist between overall spectral shapes and light-curve decline rates. We introduce a novel standardization method for SN II spectra. After empirically flattening the spectra via continuum normalization, we interpolate the resulting "feature spectra" onto a fixed grid of epochs using Gaussian Process regression. The interpolated spectra are then analyzed using Principal Component Analysis to explore correlations. We find that SNe IIP and IIL form a continuum spectroscopically, though some clustering remains. The spectral diversity is characterized mainly by two components: one continuous group with well-defined P-Cygni profiles and another with "less-regular" features likely driven by enhanced circumstellar material (CSM) interaction. Our results reveal that the spectral diversity of SNe IIP/L diminishes over time. We confirm observational correlations: steeper light-curve declines correspond to weaker spectral features, indicating that SNe IIL tend to show weaker emission and, in some cases, a lack of distinct absorption lines. These trends seemingly break down by enhanced CSM interaction that affects the P-Cygni profiles. Our data-driven method reveals underlying spectral correlations and supports a continuous distribution between IIP and IIL subtypes. This method paves the way for more refined classification algorithms.

Figures (13)

• Figure 1: Example of the LOWESS-based continuum removal applied to selected spectra of SN 2012aw. Top left: Original spectra with the LOWESS-smoothed continuum overplotted. The yellow marker indicates a reference wavelength at which the local linear fits are displayed for each spectrum. The blue shaded region highlights the masked H$\alpha$ feature, excluded from the continuum fitting to avoid bias Bottom left: Shape of the LOWESS kernel at the reference wavelength, illustrating the weighting scheme used during smoothing. Right: Continuum-normalized spectra obtained by dividing the original data by the fitted LOWESS continuum.
• Figure 2: Gaussian Process denoising applied on SN 2015bs example spectra. The left plot shows the original spectra overlaid with their denoised counterparts, demonstrating the effectiveness of the procedure in preserving spectral features. The right panel shows histograms of the flux residuals attributed to noise. The near-Gaussian distribution of these residuals supports the assumptions of white noise and confirms that no significant correlated features were removed during the process.
• Figure 3: Illustration of the Gaussian Process-based interpolation of the spectral time series. Panel (a) 3D view of the continuum-normalized spectral time series of SN 2007od interpolated onto a common time–wavelength grid. Panel (b) projection of panel (a), collapsed along the time axis, highlighting the evolution of spectral changes in the region around H$\alpha$. Panel (c) flux evolution at a given wavelength bin (red dashed line in panels (a) and (b)), with GP fit color-coded by epoch.
• Figure 4: PCA results at one representative epoch. The figure shows the PCA applied to continuum-normalized spectra at 20 days post-explosion. The top spectrum represents the mean SN II spectrum at that epoch, while the spectra below show the first few eigenspectra derived from the PCS (via singular value decomposition, SVD), ordered from top to bottom by decreasing significance. For reference, the mean spectrum is overplotted in light gray behind each eigenspectrum. The numbers next to each eigenspectrum indicate the corresponding eigenvalue, which reflects the variance explained by that PC. The inset plot in each panel displays the eigenvalue decay, indicating the variance captured by each component.
• Figure 5: t-SNE projection plots of the PCA coefficients (per epoch). These plots show how the SNe are distributed within the high-dimensional PCA space after dimensional reduction. Colored pentagons mark the positions of different cmfgen models variants. SNe 2023ixf and 2024ggi are also added using the spectra currently available in the literature (red crosses), along with CSM-interacting SNe II from Jacobson-Galan2024 (green stars). A clear separation by $s_2$ values is observed across all epochs, confirming the link between spectral and light curve properties. At 50 days, two groups emerge: a continuous "regular SNe IIP/L" sequence along $s_2$, and a scattered subgroup, likely shaped by strong CSM interaction (marked by the yellow rectangle). Model projections align with expectations: CSM-rich models explain the subgroup objects, while mass-loss models resemble fast-declining SNe.