It is very difficult to obtain the stellar parameters of Be stars. The effects of their fast rotation (line broadening, gravity darkening, oblateness) and the contamination of the disk (emission lines, IR excess) have profound effects on their observables, making it hard to analyse them with traditional methods. Obtatining disk parameters is just as hard, as the central star+disk is a coupled system, and the parameters, degenerate.

BeAtlas is, basically, a grid of radiative transfer models for Be stars, including synthetic spectra, SEDs and images covering the whole range of the Be phenomenom, including models for active (as in, currently with a disk) and inactive (diskless) Be stars. In my work, we combine the grid with an MCMC code that explores the parameter space and samples the posterior distribution of the parameters of the system. With this combination, we can find reliable estimates for the stellar and disk parameters of Be stars with unprecedented speed and ease.

Check out the paper on MNRAS

If an isolated Be star can be complicated, then putting them in a close binary system means double trouble. I am, however, very brave. So in my PhD I calculated a grid of smoothed particle hydrodynamics (SPH) models for these systems to understand what are the effects of the companion on the disk, and vice-versa.

We updated the SPH code used in previous works that focused on Be binaries to allow for particle splitting: in this way, or simulations can see much more of the less dense parts of the system. We can identify very distinct regions of the system, in particular the bridge that connects the Be disk to the companion, the accretion flow around the companion, and the circumbinary structure that hugs the whole system.

Combining these SPH simulations with the radiative transfer code HDUST, we can also obtain synthetic spectra, SED and images of these fake binaries, which can be used to plan and interpret observations of real systems.

Check out the paper on A&A

I was very fortunate to participate in the 2023 KAVLI Summer Program in Stellar Astrophysics, hosted at the Max Planck Institute for Astrophysics in Garching, Germany. 

The effectiveness and stability of mass transfer in a binary system are crucial for determining the final product of its evolution. Rapid binary population synthesis codes simplify the complex physics of mass transfer and common-envelope evolution by adopting parameterised prescriptions for the stability of mass transfer, accretion efficiency in stable mass transfer, and the efficiency of common-envelope ejection. Our goal is to calibrate these uncertain parameters by comparing binary population synthesis models with observational data. Our data: over 500 WD+MS binaties in the APOGEE-GALEX-Gaia catalog (AGGC), with well-measured radial velocities. Our simulations: COSMIC rapid-BPS. Our results: In the paper below ;) 

Check out the paper on A&A

Be stars, like all other massive stars, lose a lot of mass during their life. However, in their special case, mass loss leads to the formation of a circumstellar disk as this ejected material spreads outward viscously, resulting in characteristic observables like polarisation, IR flux excess, and emission lines. While the theory behind the viscous disk growth is well-established, the mechanism driving mass loss (the Be phenomenon) is still unclear. For many Be stars, mass loss occurs in outbursts, marked by increased brightness and Hα emission (depending on the system's inclination). 

The data show asymmetrical mass loss in all cases, with material ejected from a small region, likely near the stellar equator, and spreading outward to form the decretion disk (Labadie-Bartz et al. 2025). In this work, we conducted 3D SPH simulations of asymmetrical, localised outbursts, considering not just viscosity, but also the geometry of the injection volume and the velocity of the ejecta. By combining these simulations with the 3D radiative transfer code HDUST, we compared synthetic observables to our data. This allowed us to constrain the ejecta velocity, viscosity, and geometry required to match the observations.

Check out our results on the paper here