Research
I primarily study core-collapse supernovae across the full span of their evolution: from shock-cooling emission in the first hours to days, to cold dust condensing in the ejecta thousands of days later, and I help build the systems that make catching those moments possible.
Dust factories in ordinary core-collapse supernovae
Core-collapse supernovae are among the leading candidates for the dust that fills young galaxies, but almost everything we know observationally comes from a handful of famous objects, SN 1987A, Cassiopeia A. The obvious question is whether normal supernovae behave the same way.
With JWST/MIRI I am assembling the first uniform mid-infrared sample built to answer that: eleven nearby Type IIP supernovae, among them SN 2017eaw, SN 2020jfo, SN 2021gmj, SN 2021yja, SN 2022acko, SN 2023ixf and SN 2024ggi; observed between roughly 400 and 2300 days after explosion. Modelling their warm and cold dust components on a common footing lets us trace a dust-formation timeline rather than a single snapshot, and separate newly condensed grains from infrared echoes and circumstellar-interaction heating.
I lead the Keck and Gemini spectroscopic campaigns that support this work, searching for late-time signatures of circumstellar interaction and dust formation in the same objects, and I am a co-investigator on JWST programs watching dust condense in real time in two very nearby core-collapse supernovae.
The earliest moments after a SN explosion
When a massive star explodes, the shock breaking out of its surface produces a brief cooling phase whose brightness and colour evolution depend directly on how big the star was. Catch it, and you measure the progenitor. Miss it by two days, and the information is gone.
I led the discovery and analysis of early shock-cooling emission in the Type IIb SN 2024uwq, using rapid multi-band photometry and spectroscopy to constrain the radius and envelope mass of a partially stripped progenitor. Applied to a statistical sample, which the Rubin Observatory era will finally make possible, the same approach can tell us what fraction of stripped-envelope supernovae hold on to an extended envelope at all, and how that connects to their mass-loss history.
To get ahead of that, I am PI of Bok/90Prime programs at Steward that pilot early-transient discovery with ZTF, and lay groundwork for UV transient science with ULTRASAT.
AT 2021lwx, “Scary Barbie”
AT 2021lwx is the most energetic accretion event yet observed. At a spectroscopically measured redshift of z = 0.995 it reached a peak pseudo-bolometric luminosity of log(L/[erg s−1]) = 45.7 and stayed bright across more than 1000 observer-frame days. It is too smooth for AGN flaring, and too luminous for any supernova we have seen or modelled.
The optical spectra show strong, narrow-cored Balmer emission and semi-forbidden Si III], C III] and C II] lines, with none of the [O II] or [O III] that normally betrays an active nucleus. We concluded that this is most likely an extreme tidal disruption event: modelling the ZTF photometry with MOSFiT points to a ≈14 M☉ star disrupted by a ∼108 M☉ black hole. Remarkably, no host galaxy has been detected.
I am now PI of an HST program and a JWST Cycle 5 program (GO-11424) designed to find that host and test black-hole–host scaling relations for extreme nuclear transients, with a late-time analysis paper in preparation.
Gravitational-wave follow-up: finding the light
A LIGO–Virgo–KAGRA alert arrives with a localisation covering hundreds of square degrees, a few dozen plausible host galaxies, and a kilonova that fades in days. Turning that into an identified counterpart is a problem of triage under time pressure, and just as much a problem of knowing what the contaminants look like. Most of what you find in a localisation volume is not a kilonova.
SAGUARO
I work within SAGUARO (Searches After Gravitational waves Using ARizona Observatories), which turns Arizona's telescopes into a rapid-response network for compact-object mergers. Alongside it I am PI of dedicated MMT and Magellan programs at Steward for O4 and O4-IR1 follow-up, giving the collaboration guaranteed spectroscopic and imaging time to characterise candidates rather than merely detect them. Deciding which of thirty candidates gets the next hour on a 6.5-metre telescope, at three in the morning, is the actual job.
TROVE
Candidate vetting is where these searches succeed or fail, and it does not scale by hand. I contribute to TROVE (the Tool for Rapid Object Vetting and Examination), which ingests alert-stream candidates and surfaces the information a human needs to make a fast, defensible call: host association, archival history, cross-survey photometry, and the discriminants that separate a kilonova from a young supernova or a nuclear flare. I also coordinate through AZTEC and the LSST Discovery Alliance TVS–LVK effort, which is where this becomes a Rubin-era problem: the same alert volume, a much larger haystack.
What this has produced
The Type IIb SN 2025ulz is the clearest example. It sat in the localisation volume of the low-significance event S250818k, looked briefly like a kilonova, and turned out to be a supernova. That is the ambiguity you have to settle quickly and correctly, because a wrong call propagates through every follow-up decision that comes after it. We also searched for a counterpart to the subsolar-mass candidate S251112cm, a signal that would be hard to explain with any ordinary compact object.
Science-driven forecasts for transient follow-up
The Rubin Observatory will report millions of alerts per night against a global spectroscopic capacity that is essentially fixed. The scarce resource is not detection. It is deciding which objects deserve a telescope, and on which night.
The Recommender Engine for Intelligent Transient Tracking (REFITT) attacks that by prioritising follow-up on physical grounds. I showed that pre-computed hydrodynamical model grids combined with statistical inference can characterise a core-collapse supernova while it is still rising, recovering progenitor and explosion parameters from sparse ZTF light curves, and then identify the epochs where one extra observation actually changes the answer. That turns follow-up from first-come-first-served into something closer to an optimisation problem.
Getting ready for the Rubin era
Much of what I do is infrastructure for the next decade: building the discovery and follow-up channels that early-time and late-time transient science will both depend on. I am a co-investigator on SHADOW, the DECam survey shadowing Rubin/LSST to find the youngest transients in the nearby universe, and on PASSTA, the public AEON spectroscopic survey for transient astronomy on SOAR. I work within DLT40, the Global Supernova Project, TROVE and the ULTRASAT collaboration, and I have contributed to commissioning the new 90-inch instrument on the Bok telescope at Steward.
The first result from the LSST shadow survey, the restless luminous blue variable AT 2017des in NGC 4532, shows what shadowing buys you: a well-sampled record of a star behaving badly for years before anything definitive happens to it.