Learn more

Looking up at the sky on a starry night, we would be justified in thinking that the blackness between stars was just empty space. But in reality, the void is teeming with action.

 

The cosmos is filled with extremely low-density gas, whose temperature and chemical composition vary from place to place. In addition to gas, interstellar space contains dust that absorbs and reddens the light coming at us from stars.

 

Based on what we know from cosmological theory, the primordial gas that resulted from the big bang essentially contained only hydrogen and helium, the two lightest elements. The first stars were born from it and, once they died, they enriched it with new elements forged in their hot interior.

 

Star formation occurs when interstellar gas thickens and cools enough to collapse under its own gravity. Then interstellar clouds form, condensing into globules capable of giving birth to an entire cluster of stars. The study of these stellar nurseries is an active area of astrophysical research.

 

Gas swirls around in a turbulent fashion, making direct numerical simulation very difficult. In addition, the density contrast achieved between the dense star-forming cores and the rest of the rarified gas requires dedicated numerical approaches, such as adaptive mesh refinement. The physics of star formation also does not stop at hydrodynamics: gas photochemistry, the presence of dust, and magnetic fields further complicate the picture.

 

Very often the kind of hydrodynamic simulations that are required to model the formation of a stellar nursery require weeks or months of runtime on a dedicated computer cluster. This is Alessandro's work, which he carries out within a world-class international group, funded by a prestigious ERC grant from the European community.

 

In a way, the large investment of computational resources helps make the simulations unique: a plus for art, but a problem for science, which needs to be reproducible.

 

Mario received a Marie S.Curie fellowship (also from the EU) to solve this problem: how to create as many simulations as needed to gather statistics and scientific insight, with fewer computational resources? With artificial intelligence, of course. It is while working on this project that he stumbled upon the idea of converting simulations into astronomical images using a generative adversarial network, the type of neural network on which his project is based.