Research

---

Spectroscopy is a very important technique in astronomy. By splitting the light of a star in its different frequencies, astronomers can learn a lot about the target star: the temperature and gravity at the surface, the chemical composition, the speed at which it rotates about its axis and propeties of convective motions in its atmosphere. Because of the wealth of information in the spectrum of a star, such data is valuable, and used in many different fields in astronomy. For this reason, there exist many large databases of spectra of stars. Analyzing one of those spectra, however, takes a lot of manual work. Not only does manual work take a lot of time, it also introduces significant, but unquantifiable errors. In most astrophysical research, however, this source of uncertainty is not considered. For a detailed study of a single object, the time required to analyze a spectrum is not very relevant. For research involving hundreds or even hundreds of thousands of objects, though, any manual labor required in the analysis of data is a major limitation.

For my MSc thesis, I meant to solve this problem by developing a fully automated pipeline that analyzes spectra of stars and provides properties of the stars, without any manual intervention. Additionally, the pipeline had to be fast. It was meant to analyze millions of spectra, and nobody wants to use a pipeline that takes thousands of hours on a supercomputer to finish. The result of this work was HermesNet: a pipeline that did just that, making use of neural networks to drastically speed up the process. HermesNet is made for spectra of the Hermes instrument on the Mercator telescope at the Roque de los Muchachos Observatory.

With what I learned from developing HermesNet, I joined up with my friend Ragnar Van den Broeck – who developed a pipeline similar to HermesNet using different methods – to develop the next generation of our pipelines: PASTA ("Pipeline for Automated Spectroscopic Analysis"). PASTA is capable of estimating 5 properties from a spectrum: the effective temperature (which can be thought of as the surface temperature), the gravity at the surface, the metallicity (the amount of elements that are heavier than helium), the rotation speed and the radial velocity of the star (how fast the star is moving away or towards us). It is slightly slower than HermesNet, but much more accurate, and it also works on spectra of any instrument. PASTA is the first pipeline of its kind, and will empower astronomers for the first time to analyze spectra of millions of stars in a consistent way. We're currently working on implementing PASTA in the data reduction pipeline of the Sloan Digital Sky Survey (SDSS-V).

The interstellar medium is everything in-between the stars. You might think interstellar space is just an empty void, but quite the opposite is true! There are massive clouds of gas and dust, the densest of which are the birthplaces of new stars. In these interstellar clouds, a lot of chemistry is going on. For my PhD, I study the chemistry of large carbon-based molecules (fullerenes) in those places. The most common places to find these molecules are in the vicinity of new-born massive stars and dying Sun-like stars, in so-called "photo-dissociation regions" (PDRs). Coincidentally, these places also provide the most beautiful pictures you'll see from space! I am currently studying a planetary nebula – the remnant of a dying Sun-like star – with observations from the shiny new James Webb Space Telescope, which is super exciting!