research

ALMA, JWST, and extreme adaptive optics now resolve planet-forming disks in remarkable detail. Rings and gaps are ubiquitous, and some disks also show spirals, shadows, and warps. My research asks what these features tell us about the planets forming inside them and about the physics of the disks themselves. Click on a theme below for highlights, movies, and key papers.

Shadow-driven dynamics

How shadows cast by inner disks launch spirals, drive accretion, and warp the outer disk

Young planets in disk surveys

Inferring the hidden planet population from disk surveys

Turbulence and kinematics from radiation hydrodynamics

The disk's thermal structure sets where the vertical shear instability operates

Probing disk cooling with time-dependent features

Measuring cooling times and fundamental disk properties, such as gas density and grain size, from how dust and gas respond to shadows

Dust properties from multi-wavelength observations

Measuring dust size and porosity with continuum and polarization

Machine learning and AI-assisted inference

From neural networks that weigh hidden planets to AI-accelerated models of disk kinematics and Gaia images

Future directions

My program has two connected pillars. The goal is to tell apart disk substructures carved by planets from those with other origins, and to build a quantitative, predictive link between planet properties and what we observe.

I. Multi-physics models. Radiation, gas, dust, and planets should evolve together rather than treating radiative transfer as static post-processing. I am among the first users of PASTA, a next-generation GPU-accelerated code developed by my close collaborator Yan-Fei Jiang, and I help test it. PASTA now includes a dust Boltzmann treatment, in which dust is modeled as a kinetic component whose streams can cross rather than as a pressureless fluid, together with dust growth, coagulation, and coupling to N-body dynamics. My focus is on exploring all of these aspects in protoplanetary disks and other systems. Because dust sets most of the disk opacity, evolving dust and radiation together is essential for predictive models. Modern GPU architectures make this high-dimensional problem feasible and open physical regimes that were previously out of reach, especially in the inner disk, where terrestrial planets form.

About PASTA (publicly available soon, after thorough testing)

• Radiation transport: explicit and implicit multi-group radiation with implicit radiation–matter coupling.
• Dust: a kinetic (Boltzmann) dust treatment that lets dust streams cross, with dust growth and coagulation.

Shadows, accretion bursts like those in DQ Tau, dust growth, and vertical structure are natural experiments for these models. Chemistry is the next step.

II. AI-assisted inference across images, spectra, and time. Building on PGNets, I plan to develop simulation-based inference that first identifies which physical process made a structure and then measures its properties, and to use machine-learning surrogates on modern GPU/CPU architectures to accelerate Discminer modeling of ALMA kinematics. I am also building software to measure shadow motions across a population of disks with Gaia DR4 residual images (expected December 2026). See machine learning for details.