🔭 astrophysics

Traces of Helium Detected in Type Ic Supernova 2014L

By applying a deep-learning-accelerated Bayesian radiative transfer analysis to the optical and near-infrared spectra of Type Ic supernova 2014L, this study provides robust evidence for the presence of approximately 0.018–0.020 solar masses of helium in the ejecta, challenging the conventional view that helium is absent in such events.

Original authors: Jing Lu, Wolfgang E. Kerzendorf, John T. O'Brien, Maryam Modjaz, Jared A. Goldberg, Nutan Chen, Erin Visser, Joshua V. Shields, Andrew G. Fullard

Published 2026-04-02

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Jing Lu, Wolfgang E. Kerzendorf, John T. O'Brien, Maryam Modjaz, Jared A. Goldberg, Nutan Chen, Erin Visser, Joshua V. Shields, Andrew G. Fullard

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Mystery: The "Ghost" Helium

Imagine a supernova (a massive exploding star) as a giant, glowing firework. Astronomers classify these fireworks based on the "smoke" they leave behind. Type Ic supernovae are the ones that, according to the rulebook, should have absolutely no helium left in their outer layers. It's like a cake that is supposed to be chocolate but has no cocoa powder in the recipe.

For decades, scientists have argued: Is the helium actually gone, or is it just hiding?

The Optical View: When we look at the explosion with standard telescopes (visible light), the helium is invisible. It's like trying to find a specific type of fish in a murky pond using only a flashlight; the water is too cloudy, and the fish blends in with the other debris.
The Infrared View: The authors of this paper decided to look through a different "lens"—near-infrared light. This is like switching from a flashlight to a thermal camera. Suddenly, the "ghost" helium becomes visible because it glows in a way that other elements don't.

The Detective Work: How They Solved It

The team studied a specific explosion called SN 2014L. To figure out what was inside, they didn't just guess; they built a digital "time machine."

The Super-Computer Simulator (TARDIS): They used a powerful code called TARDIS to simulate what a supernova explosion should look like if it had different amounts of helium, carbon, oxygen, etc. Think of this as a video game engine that renders realistic explosions.
The AI Speed-Runner (The Emulator): Running the simulation takes a long time (like waiting hours for a movie to render). To speed things up, they trained a Deep Learning AI (a neural network) to act as a "cheat code." This AI learned the patterns of the slow simulation and could predict the results in a split second. It was like having a super-fast assistant who memorized the entire physics textbook and could answer questions instantly.
The Bayesian Detective: They used a statistical method called Bayesian Inference. Imagine you are trying to guess the ingredients of a soup by tasting it. You start with a guess, taste the soup, and then adjust your guess. You do this millions of times until you are 99% sure of the recipe. The team did this with light spectra, comparing their AI predictions to the actual light from SN 2014L.

The Big Discovery: Helium Was There All Along

After running millions of simulations, the results were clear:

The Verdict: SN 2014L does have helium. It's not a lot (about 2% of the mass of our Sun), but it is definitely there.
The Smoking Gun: The "smoking gun" was a dark spot in the near-infrared light at a specific wavelength (1 micron).
- When the team ran the simulation without helium, this dark spot disappeared.
- When they added a tiny bit of helium, the dark spot appeared exactly where it was seen in the real data.
- The Analogy: It's like trying to identify a singer in a choir. If you remove the bass singer, the low notes vanish. The team realized the "low notes" in the supernova's light could only be made by helium.

What Else Did They Find?

The Density: They figured out how the star's material was spread out. It wasn't random; it followed a specific mathematical curve (a power law) that matches what physics predicts for a star that is dominated by radiation pressure. It's like the debris from an explosion spreading out in a perfect, predictable cone.
The Carbon: The outer layers were surprisingly rich in Carbon. This suggests the star was a "Carbon-rich" star before it died, which helps scientists understand how massive stars evolve and lose their outer layers.
The Calcium Trick: They found that Calcium (which makes bright lines in the light) was actually quite rare in the outer layers, but because of how the star was expanding, it still looked bright. It's a reminder that just because something looks bright, it doesn't mean there's a lot of it there.

Why Does This Matter?

This paper changes the rulebook.

Binary Stars: Many Type Ic supernovae are thought to come from stars in a "dance" with a partner star that steals their outer layers. If the partner steals all the helium, we shouldn't see any. But since we do see a little bit, it suggests the "stealing" process isn't 100% efficient, or the stars are losing mass in a different way than we thought.
The Power of Infrared: This study proves that to understand the true composition of these cosmic explosions, we must look in the infrared. Relying only on visible light is like trying to read a book in the dark; you miss the most important details.

In a nutshell: The team used a super-fast AI and a "thermal camera" to prove that a star explosion thought to be helium-free actually had a small, hidden amount of helium. This solves a decades-old mystery and helps us understand how massive stars die.

1. Problem Statement

Type Ic supernovae (SNe Ic) are defined by the absence of hydrogen and helium features in their optical spectra, implying their progenitors have lost their outer H/He envelopes. However, a long-standing debate exists regarding whether helium is truly absent or merely spectroscopically undetectable in the optical band due to low ionization states or blending with other ions.

The Challenge: Helium lines in the optical are often weak or blended. The Near-Infrared (NIR) region contains cleaner He lines (e.g., He I 1.083 µm and 2.058 µm), but these are often attributed to blends of other ions (C I, Si I, Mg II) in previous studies.
The Gap: Previous constraints on Helium mass relied on limited model sets and direct radiative-transfer simulations, which are computationally too expensive to perform the millions of evaluations required for robust Bayesian inference across a wide parameter space.

2. Methodology

The authors employed a Bayesian inference framework accelerated by a deep-learning emulator to perform full spectral modeling of SN 2014L.

Observational Data:
- Optical Spectrum: Obtained at peak light (MJD 56693.9, 0 days relative to peak) using the YFOSC instrument on the 2.4m Li-Jiang Telescope.
- NIR Spectrum: Obtained 2 days post-peak (MJD 56695.9) using the FIRE instrument on the 6.5m Magellan Baade Telescope.
- Data covered the wavelength range 3500 Å to 24,000 Å.
Radiative Transfer Code (TARDIS):
- Used the open-source code tardis with Non-Local Thermodynamic Equilibrium (NLTE) treatments for Helium (specifically the analytical approximation by Boyle et al. 2017).
- Ejecta Model: Parameterized the outer ejecta (above the photosphere) with a power-law density profile ( $\rho \propto v^{\alpha_\rho}$ ) and a uniform composition for most elements. Calcium (Ca) was modeled with a two-zone structure (depleted in the outer zone) to match observed high-velocity features.
- Parameters: 18 free parameters including luminosity ( $L$ ), time since explosion ( $t$ ), density exponent ( $\alpha_\rho$ ), transition velocity for Ca, and mass fractions for 13 elements (He, C, O, Ne, Na, Mg, Si, S, Ca, Ti, Cr, Fe, $^{56}$ Ni).
Deep Learning Emulator:
- To overcome the computational bottleneck of running tardis millions of times, the authors trained a Probabilistic Dalek neural network emulator.
- Training: Generated $\sim$ 700,000 synthetic spectra using tardis. The emulator achieved a mean fractional error of 1% compared to tardis outputs across all wavelengths.
- Inference: The emulator was integrated into the ultranest nested sampling algorithm to explore the posterior distribution of the ejecta properties.

3. Key Contributions

First Quantitative Bayesian Constraint on He in SN Ic: This is the first study to use a probabilistic emulator to rigorously constrain Helium mass in an SN Ic by jointly analyzing optical and NIR spectra.
Demonstration of NIR Necessity: The study proves that optical spectra alone are insufficient to detect trace Helium; the NIR region is critical for breaking degeneracies and identifying He I features.
Computational Efficiency: Successfully demonstrated that deep-learning emulators can replace direct radiative transfer simulations in Bayesian inference, reducing computation time from impractical (years) to feasible (24 hours) while maintaining physical fidelity.
Refined Density and Composition Models: Provided precise constraints on the density power-law exponent and elemental mass ratios, moving beyond simple upper limits to specific posterior distributions.

4. Key Results

Detection of Helium:
- The observed spectra are statistically inconsistent with a model containing zero Helium.
- The inferred Helium mass above the photosphere ( $v_{inner} \approx 9164$ km/s) at peak light is 0.018 to 0.020 $M_\odot$ (16%–84% credible interval).
- Mechanism: The strong absorption feature near 1.083 µm is dominated by He I (contributing ~70% of the packet interactions in the 1.00–1.07 µm range), with partial blending from C I. The weak He I 2.058 µm feature is also reproduced.
- Models without Helium failed to reproduce the 1 µm NIR feature, even though optical features remained similar.
Density Structure:
- The density power-law exponent is constrained to $\alpha_\rho = -7.04$ to $-6.88$. This is consistent with theoretical predictions for radiation-dominated explosions ( $\alpha_\rho \approx -7$ ).
Chemical Composition:
- Carbon-Rich: The outer ejecta is dominated by Carbon ( $\sim$ 1.25 $M_\odot$ ), with a high C/O mass ratio ( $\sim$ 13).
- Calcium Stratification: A uniform Ca abundance produced unrealistic high-velocity features. The best-fit model requires a transition velocity ( $v_{X_{Ca}=0}$ ) of 18,162 to 18,281 km/s, above which Ca is depleted. This explains the observed Ca II features without requiring high Ca abundance in the outermost layers.
- Iron Group Elements (IGE): The mass of IGEs above the photosphere is low ( $\sim$ 0.0028 $M_\odot$ ).
- Neon: Remained unconstrained due to a lack of distinct spectral features.
Luminosity and Time:
- The inferred peak luminosity is $\sim$ 2 times higher than previous estimates, suggesting stronger host extinction than previously thought.
- The time since explosion at peak is inferred to be $\sim$ 35 days, potentially longer than the 13-day rise time estimated from light curves, possibly due to non-homologous expansion phases or central engine energy injection.

5. Significance

Progenitor Evolution: The detection of trace Helium ( $\sim$ 0.02 $M_\odot$ ) challenges the notion that SNe Ic are completely He-free. It suggests that binary stripping or stellar winds may not remove all Helium, or that the remaining He is spectroscopically hidden in the optical but visible in the NIR.
Methodological Advancement: This work establishes a new standard for supernova analysis. By combining radiative transfer with Bayesian inference and deep learning, it allows for systematic, uncertainty-quantified constraints on ejecta properties that were previously impossible to obtain.
Future Applications: The framework is scalable to larger samples of SNe Ib and Ic, enabling systematic studies of residual Helium, C/O ratios, and the efficiency of envelope stripping mechanisms in massive star evolution.

In conclusion, the paper resolves the "Helium Question" for SN 2014L by demonstrating that while Helium is scarce, it is definitively present and spectroscopically detectable in the NIR, necessitating a revision of the "He-free" classification criteria for some Type Ic supernovae.