Mark Magee
I am a Warwick Prize Fellow working in the Astronomy and Astrophysics Group at the University of Warwick. I am a member of the Extragalactic Transients Group and my research focuses mainly on the study of supernovae with a particular emphasis on thermonuclear supernovae.
Many stars in the Universe end their lives as white dwarfs and some of those white dwarfs in binary systems will eventually explode as thermonuclear supernovae. Exactly how and why this occurs however remains a mystery, despite the hugely important role thermonuclear supernovae play in many areas of astrophysics. Although thermonuclear supernovae are often considered to be a homogeneous class, more and more diversity is being discovered that only adds to the mystery.
My work includes trying to understand which types of binary systems produce thermonuclear supernovae and the different explosion mechanisms responsible for the supernovae we observe. To do this, I use a combination of observational and theoretical modelling studies. Observationally, my work includes detailed studies of individual objects with extensive data sets and larger studies centred around the properties of specific supernova populations. My theoretical work includes developing sophisticated models for different explosion scenarios that can be be directly tested against observations. I am interested in developing new machine learning techniques to overcome current limitations associated with our analysis techniques and the era of `big data'.
Accelerating spectroscopic modelling with machine learning
Type Ia supernovae (SNe Ia) are thermonuclear explosions of white dwarfs in binary systems. Their standardisable light curves make them extremely important distance indicators in cosmology. Fundamentally however we still don't fully understand why SNe Ia explode or indeed why they are standardisable. Improving our knowledge of SNe Ia physics is key to driving down systematic uncertainties that limit the precision of cosmological parameters measured from SNe Ia light curves and hence our understanding of the Universe. Comparisons of SNe Ia observations with explosion models are the best hope of resolving this long-standing issue, but this presents its own challenges. Highly sophisticated, multi-dimensional explosion simulations allow us to directly compare against observations although the high computational expense associated with generating such simulations generally limits them to only a relatively small number of simulations.
Other tools, such as TARDISLink opens in a new window, allow for more rapid generation of models. TARDIS is a radiative transfer code designed for modelling spectra of supernovae. I am a member of the TARDIS collaboration and core team. Using TARDIS, one can develop a more simplified model of the SN ejecta, generate spectra, and compare against observations. By applying this method to multiple spectra at different epochs, one can build up a picture of the ejecta structure to better understand properties of the explosion. There are some limitations to this approach however as such comparisons have typically been done using qualitative means so far.
In Magee et al. 2024Link opens in a new window I present a method of accelerating this analysis and making it more quantitative, which I call riddler. Using TARDIS, I generated a large number (~100,000) of model spectra for two explosion simulations and used these as training data for neural network emulators. Although TARDIS simulations are relatively fast (~10+ minutes), they are sufficiently long that covering a large parameter space quantitatively is infeasible. The emulators however are able to reproduce the TARDIS simulations in a fraction of the time, making it possible to generate thousands of spectra in an instant. These can then be compared against observations to determine the best-matching explosion model. Magee et al. 2024 demonstrated the utility of this approach, but was limited to only two explosion models. Work is currently underway expanding this to a more representative sample covering the full breadth of SNe Ia diversity. It is clear that such rapid and quantitative modelling of SNe will become increasingly important as spectroscopic samples continue to grow larger.
An example of the neural network architecture used by riddler
Further reading:
The earliest moments of thermonuclear supernovae
Observations within the first few days of explosion hold key diagnostic clues as to the origin of the type Ia supernovae (SNe Ia). The light curves of SNe Ia are powered by the radioactive decay of 56Ni produced during the explosion. Changing where that 56Ni is distributed within the ejecta however can have a significant impact on the overall shape of the light curve. Within these early phases, many SNe Ia have also been observed to show 'bumps' in their light curves. These bumps are short-lived and typically only last up to a few days after explosion, making them difficult to find and study.
A few different scenarios have been proposed to explain these bumps, but they generally fall into two categories: interaction and radioactive decay. For interaction-based scenarios, the expanding ejecta released from the explosion slams into either the companion star or pre-existing material around the white dwarf. In either case, the ejecta becomes shock-heated and emits a burst of light at X-ray, UV, and optical wavelengths lasting up to a few days depending on the properties of the system. Conversely, some explosion scenarios (such as the double detonation scenario) predict significant amounts of short-lived, radioactive isotopes in the outermost regions of the SN ejecta. Just as radioactive decay of 56Ni powers the main light curve, decay of these isotopes can produce a bump within the first few days after explosion. Clearly, the early light curves of SNe Ia can provide significant constraints on the explosion physics and progenitor system.
Examples of early light curve shapes produced by different mechanisms
To improve our understanding of these early phases, I developed a radiative transfer code, TURTLSLink opens in a new window, designed to model thermonuclear supernovae shortly after explosion and up to approximately maximum light. I have been systematically exploring these various scenarios and producing models that are made available to the community. In Magee et al. 2020Link opens in a new window I showed that differences in the 56Ni distribution can explain much of the observed variation among SNe Ia light curves at early phases, but existing explosion models struggle to capture this behaviour. Magee & Maguire 2020Link opens in a new window and Magee et al. 2021Link opens in a new window both explore different scenarios invoking short-lived radioactive isotopes in the outer ejecta as a means to explain early light curve bumps. The former focuses on clumps of 56Ni, which decays with a half-life of approximately a week. The latter focuses on the double detonation scenario, which is predicted to produce isotopes that decay with significantly shorter half-lives of minutes to days. Based on these models, I show how large amounts of radioactive material in the outermost ejecta can explain the shapes of some bumps, but cannot reproduce the observations at later times due to the significant line blanketing it creates. In Magee et al. 2022aLink opens in a new window, I used simulations of all-sky surveys to present the first estimate of the intrinsic rate of SNe Ia with bumps and found that they could occur in ~30% of SNe Ia. Further work is needed to provide tighter constraints and fully understand the mechanisms producing these fleeting signatures.
Further reading:
Extreme thermonuclear explosions
Although so-called normal type Ia supernovae (SNe Ia) are the most common class of thermonuclear SN observed, a wealth of peculiar and unusual events have also recently been discovered. These SNe show some features in common with SNe Ia, but are sufficiently different in other ways as to warrant a distinct classification. The most common of these are known as type Iax supernovae (SNe Iax). SNe Iax are extreme thermonuclear explosions. They show a wide range in luminosities, from slightly fainter than normal SNe Ia to over 100x fainter – we may not have even discovered the lower limit! They also show a range of light curve shapes, but generally evolve faster than normal SNe Ia, and peculiar spectral features, including low velocities and the presence of iron lines at all epochs.
Understanding extreme events such as SNe Iax can help to constrain thermonuclear explosion physics, as proposed models must explain all of the unusual features observed. One of the most promising models is a weak explosion of a Chandrasekhar mass white dwarf, known as a pure deflagration. Due to the weak explosion this scenario typically predicts faint and fast-evolving transients. In some cases the explosion is sufficiently weak that it doesn't completely destroy the white dwarf and a bound remnant is left behind.
Comparison between a SN Iax (black) and pure deflagration models with different explosion strengths (colours)
In Magee et al. 2016Link opens in a new window I presented observations of one of the best-observed SNe Iax and showed that its properties were generally consistent with predictions from pure deflagration explosion models. Another key prediction of this scenario is that the ejecta is well-mixed – the turbulent flame or deflagration results in burned and unburned material being present essentially throughout the ejecta. Some studies have argued that SNe Iax instead show evidence of a layered ejecta structure, rather than mixing, and therefore are inconsistent with the pure deflagration scenario. I investigated the signatures of layering in Magee et al. 2022bLink opens in a new window and showed that SNe Iax are indeed consistent with a well-mixed ejecta structure. Additional observations at earlier times could help to determine whether this also holds true for the very outermost layers.
Aside from the explosion mechanism, it is also important to understand the progenitor system that produces SNe Iax. Previous studies have claimed some SNe Iax show strong features due to helium. This has not been observed for other thermonuclear supernovae and helps to constrain the progenitor as helium is not produced during the explosion. In Magee et al. 2019Link opens in a new window, I investigated all SNe Iax with infrared spectra (the strongest helium absorption feature is found at these wavelengths) and those specifically claimed to show helium. I showed that these features are consistent with helium, but require a lot of helium in the ejecta and this appears inconsistent with material stripping from a companion. Other SNe Iax showed no evidence in favour of helium features.
Further reading:
The hunt for gravitationally lensed supernovae
When a massive foreground galaxy is well-aligned with a background source, the mass of the galaxy can warp spacetime to such an extent that multiple images of the background source can be observed. This phenomenon is known as strong gravitational lensing and is exceedingly rare. In addition to forming multiple images, the background source can also be magnified. Gravitational lensing therefore allows us to see more distant objects that otherwise would have been too faint to observe.
If the background source is time-varying, such as a supernova, we can measure the arrival times between the different lensed images. As these images have taken different paths to reach us, we can use these time-delays to measure the expansion rate of the Universe. This makes gravitationally lensed supernovae (glSNe) a powerful probe of cosmology. Unfortunately however only a few glSNe are known. Larger samples of glSNe with precise time delay measurements are needed to provide competitive constraints on the expansion rate.
Finding lensed supernovae is challenging. In Magee et al. 2023Link opens in a new window, I led the first archival search for glSNe within the Zwicky Transient Facility (ZTF) public survey. Using methods proposed within the literature, I searched through ~30,000 transients for signatures that could be indicative of glSNe, including extremely red colours (due to the high redshift) and bright absolute magnitudes (due to the lensing magnification and mis-identified redshifts). Unfortunately, I found no strong evidence in favour of any glSNe. While disappointing, this is not altogether unsurprising. Using simulations, I showed that among these ~30,000 transients we would expect up to one real glSN. Our current methods of identifying glSNe are limited for shallow surveys, such as ZTF, therefore I propose additional selection criteria that should aid in the identification of lensed supernovae with upcoming facilities.
Selecting glSNe based on red colours produces many false positives, but also misses a lot of real glSNe
Further reading:
