Disk images by ESO; Adapted by Olena Shmahalo/Quanta Magazine
Research
Protoplanetary disks are rotating sheets of gas and dust around newborn stars. They are the birthplaces of planets. Over a few million years, a variety of physical and chemical processes take place in these disks, giving rise to planets that eventually carve gaps, rings, and spirals. Our group studies every aspect of this transformation. We run numerical simulations to model disks, capture high-resolution images to characterize disk structures, and build machine-learning tools to replace full-physics calculations. Together, these methods reveal how disks evolve, how planets form, and why planetary systems differ so widely.
An excellent introduction to our group's work by Adam Hadhazy at the Kavli Foundation. More media reports on our work can be found under the tab 'Media Exposures'.
Sample our research in simulations, observations, and machine learning below.
Simulations
We run hydrodynamic and radiative-transfer simulations using state-of-the-art tools that follow disks from their turbulent beginnings to the moment planets sculpt gaps, rings, and spirals. These numerical experiments let us study how planets form and how they interact with their birth environments.
The movie on the right shows how a super-Earth planet can create multiple dust rings and gaps in a disk with low viscosity (Dong, Li, Chiang & Li, 2017)
Articles introducing our work in plain language:
Observations
We use some of the sharpest eyes in astronomy, including ALMA, JWST, Subaru, Gemini, and VLT, to image protoplanetary disks at sub-arcsec resolution. These observations reveal rings, gaps, spirals, and crescents in exquisite detail. By comparing the data with models, we trace the composition, kinematics, temperature, and spatial distribution of disk material, building a direct bridge between disk physics and planet-formation activity.
The movie on the left shows a 3D view of wiggles in position–velocity space for the AB Aur disk in ALMA 13CO emission (Speedie et al. 2024).
Machine Learning
To accelerate discovery, we develop machine-learning models that emulate or extend full-physics calculations. Neural networks reproduce complex simulations in a fraction of the time, while computer-vision tools extract subtle features from telescope images. These data-driven methods let us explore vast parameter spaces, interpret observations in real time, and uncover patterns that traditional approaches might miss.
The movie on the right shows the predictions generated by a neural network of a disk's response to the presence of a planet (Mao et al. 2023)
