Research
Four interlocking research programmes, unified by a common challenge: recovering hidden structure from surface wave measurements.
01
Helioseismology
The Sun
The Sun rings like a bell. Millions of acoustic modes propagate through its interior, each one carrying a fingerprint of the medium it traversed. For thirty years, helioseismology has exploited this fact to measure the Sun's internal rotation, sound-speed profile, and helium abundance — but the field hit a wall. Convective velocities predicted by models are an order of magnitude larger than what observations reveal.
Our work focuses on resolving this 'convection conundrum' using deep inversion methods and convolutional neural networks trained on simulated solar oscillations. Recent results — published in Nature Astronomy — identified global-scale Rossby waves and magnetically modified oscillation modes that earlier analyses had overlooked.
Evidence for global-scale magnetically modified Rossby waves in the Sun
Nature Astronomy, 2026
02
Asteroseismology
Stars
The Kepler and TESS space telescopes have produced photometric time series for hundreds of thousands of stars with a precision no ground-based facility can match. Buried in those light curves are the frequencies of stellar oscillations — asteroseismic 'heartbeats' that encode stellar age, mass, helium content, and internal rotation.
We apply encoder–decoder neural networks and Bayesian inference to extract those signals at scale. A particular focus is red giants, where the coupling of gravity and pressure modes produces mixed modes that probe the stellar core directly. Our 2025 detection of anomalously fast core rotation in a sample of red giants challenged standard stellar evolution models.
Anomalously fast core and envelope rotation in red giants
ApJ Letters, 2025
03
Transit analysis
Exoplanets
When a planet passes in front of its host star, the resulting photometric dip — a transit light curve — encodes the planet's radius, orbital inclination, and limb-darkening profile of the star. With enough precision, repeated transits reveal atmospheric absorption, orbital decay, and transit-timing variations caused by companion planets.
We develop ML-assisted pipelines for radial velocity analysis and transit detrending that reduce sensitivity to stellar activity — the dominant noise source in exoplanet detection. Our 2026 radial velocity paper introduced a multimodal spectral model that improves planet mass constraints without requiring simultaneous photometry.
Machine learning for radial velocity analysis of stellar spectra
ApJ, 2026
04
Geophysics
Earth
The mathematical machinery of helioseismic inversions — finite-frequency sensitivity kernels, Born approximations, iterative back-projection — translates directly to seismic tomography of Earth's mantle and core. Applying methods developed for stellar physics to the solid Earth opens productive cross-disciplinary conversations.
Our geophysics work focuses on normal-mode coupling and adjoint inversions for deep Earth structure. The collaboration brings together geodesists and solar physicists, and has produced shared tooling that benefits both communities.
Full record
All peer-reviewed publications