Spectroscopie van zware sterren bij lage metalliciteit: Een gedetailleerde studie van de atmosfeer en de winden van B-sterren

Massive stars have an initial mass above 8M . Even though massive stars are less numerous than low-mass stars, they shape the galaxies they inhabit both physically and chemically. They are partly responsible for enriching their galaxy with heavier elements which are created in their core and expelled through strong stellar winds. When the fuel in the core has been exhausted they end their lives in energetic supernova explosions which also contribute to chemical enrichment of the ambient environment. The strong outflows and high-pressure ionised regions that surround massive stars in the phases of their lives where they have high surface temperatures set in motion and sweep up interstellar material, potentially halting or initialising star formation elsewhere in their host galaxies. To understand all these feedback processes caused by massive stars, it is important to understand how these stars change over their lifetime. Massive star evolution is heavily impacted by their strong mass-loss rate causing the star to loose multiple solar masses worth of material over its lifetime. In stellar evolution models mass-loss rates are included as prescriptions, simple formulas based on hydrodynamical predictions which compute the mass-loss rate from stellar parameters. There are multiple prescriptions which agree relatively well in the O- star regime (T eff > 30 kK), but disagree by orders of magnitude in the B-star regime 10 kK< Teff < 30 kK. This is partly because of a sudden increase in mass-loss rate for stars with T eff < 25 kK in some prescriptions called the bi-stability jump. It is possible to compare these prescriptions to observations by fitting the spectra of the star to model spectra based on 1D stellar atmosphere and wind models. Although some studies have compared these prescriptions to observations, they were not able to take into account the inhomogeneities of these winds, resulting in large uncertainties. This thesis has two goals: (1) empirically derive the mass-loss rate of B-stars while accounting for the inhomogeneities to assess how the mass-loss rate is influenced by different stellar parameters, coincidently also assessing the viability of the bi-stability jump; (2) improve the methods for deriving the stellar and wind parameters both in terms of the fitting routine and including more physics in the 1D spectral modelling tools. Over the course of the thesis we will show the results of the spectral fits of 41 stars in the LMC and SMC, obtaining stellar parameters, mass-loss rates, clumping parameters describing the inhomogeneities in the wind and other wind parameters. The derived mass-loss rates in both the SMC and LMC show only a weak dependence on T eff and no increase in mass-loss rate for cool B-stars disproving the bi-stability jump. As the standard mass-loss prescription in evolutionary models includes a bi-stability jump, evolutionary models overestimate the empirical mass-loss rate of stars cooler than 25 kK by up to a factor 1000 and on average a factor 30. By comparing the empirical mass-loss rates of the LMC to the SMC, we found that the metallicity dependence of the mass-loss rate is much weaker than prescriptions suggest. As iron-group metals are historically believed to be the fundamental driver of the outflows of these stars, this is surprising and warrants a closer theoretical study. The clumping description currently implemented in fastwind assumes a two-component medium, i.e. a medium having dense and rarefied regions. This description assumes that the dense regions contribute nearly all wind mass, in line with the theoretical 1D description of line de-shadowing instability. Instead, we find a wind structure where 40% of the wind mass is located in the rarefied regions, contradicting the assumption made in the models. This result is enforced by the study of multi-D radiation hydrodynamical models of stars, which find a wind-density distribution similar to our results, not distributed as a two-component medium, but rather turbulent and log-normal. This inappropriate formulation of clumping behaviour in the 1D model may explain the high uncertainties we find on the derived clumping parameters. The average density and temperature atmosphere profile of multi-D radiation hydrodynamical models do not agree with the 1D models. This is alleviated by including a turbulent pressure term into the stellar structure calculations of the 1D models. Changing the stellar structure causes a shifts in most of the derived spectral lines, notably narrowing the Balmer-lines requiring a higher stellar mass to retrieve the same shape. This mass increase may help alleviate a persistent underestimation of stellar mass estimates based on spectroscopically derived gravities compared to mass estimates from other methods. To test this theory, we implement turbulent pressure in fastwind and fit the spectra of two gravitationally bound stars for which the masses are known and previous spectroscopically derived masses are underestimating the actual mass. This new method finds spectroscopic masses in agreement with those from other methods. Using our current fitting method, each spectral fit of a star requires around 1.5 days. As the newest generation of spectroscopes will increase the amount of spectra to be analysed by orders of magnitude (e.g. 4MOST/VLT, WEAVE/WHT, MOSAIC/ELT), we also investigate new methods which could speed up our current process. The application of neural networks which are able to emulate fastwind are a possible solution. The neural network takes the same stellar and wind parameters as fastwind and outputs spectra mostly indistinguishable from fastwind spectra in a fraction of a second instead of 45 minutes.