The AI scrutinizes the mysteries of the first stars

The first stars could not have been born or have been exactly as do and are the stars that we observe today in the Milky Way. According to a team of researchers from the Kavli Institute for Physics and Mathematics of the Universe (Kavli IPMU) in Japan, one of the scenarios of their birth has left traces in the oldest stars in the Galaxy. revealed learning automatique (machine learning in English).

The Planck satellite has allowed us to observe and study as never before the oldest light in the cosmos, fossil radiation, emitted around 380,000 years after the Big Bang. This radiation confirms the theory of cold dark matter which allows galaxies to form very quickly thereafter, without which it would not yet exist today although the question is not yet settled if we consider instead of dark matter a modification of the theory of gravitation within the framework of Mond.

Between the birth of the fossil radiation and the appearance of the first galaxies there is the formation of the first stars by gravitational collapse of clouds of hydrogen and helium containing some traces of lithium, but at the beginning no heavy elements like oxygen, carbon and nitrogen of the cells of our body. All these heavy elements are called by astrophysicists “metals”. They will be produced by successive generations of stars, notably with thermonuclear fusion reactions and finally explosions in the form of supernovae.

The new nuclei produced in this way will be injected into the interstellar medium where matter continues to collapse giving rise to new stars which in turn will produce new nuclei. There is therefore a whole cycle that causes cosmic matter to evolve chemically.

For about 13.8 billion years, the Universe has been constantly evolving. Contrary to what our eyes tell us when we contemplate the sky, what composes it is far from being static. Physicists have observations at different ages of the Universe and carry out simulations in which they replay its formation and its evolution. It would seem that dark matter has played a big role since the beginning of the Universe until the formation of the large structures observed today. © CEA Research

Star populations and the history of the cosmos

Nowadays, the existence of silicate and carbonaceous dust influences the formation of stars which they facilitate. For given pressure and temperature conditions, a cloud of matter will then collapse and fragment giving clusters of young stars, often in the initial form of binary and sometimes multiple systems, such as Alcor and Mizar which form a triple system. binary.

But as this dust did not yet exist after the emission of the fossil radiation, the first stars could not form according to a scenario which has been repeated for at least 13 billion years. There are theoretical models and numerical simulations of the birth of the first stars and we try to constrain them by indirect observations.

This is precisely what a team of researchers affiliated with the Kavli Institute for the Physics and Mathematics of the Universe (Kavli IPMU) in Japan has done. They have just published the results of their work in the famous journal The Astrophysical Journal in the form of an article, an open-access version of which can be found at arXiv.

Astrophysicists and cosmologists have been accustomed to speaking of three populations of stars since the classification of those of the Milky Way in 1944 by Walter Baade. There is population I enriched in “metals” and whose stars are at most 10 billion years old. The Sun is one of them.

Then there is population II, the oldest, which contains the stars with the poorest metals. Population III has disappeared, they were the first generation of stars and they are thought to have been massive, even very massive, to the point of possibly forming the seeds of supermassive black holes. The theory of stellar evolution tells us that they must therefore have evolved very quickly, living less than a million years before exploding in supernovae after having synthesized the first heavy elements.

It was thought until now that the fragmentation of the primordial clouds giving the first stars of population III gave isolated stars and that the first stars of population II, with in particular the poorest in metals that astrophysicists call ” extremely metal-poor stars », EMPs, were therefore born in the supernova remnants of a single population III star. But recent numerical simulations suggested that in fact these stars were already forming clusters, certainly smaller and less rich in stars than those observed today in the form of stellar nurseries.

Stellar archeology leaves traces in cosmochemistry

The EMPs were therefore to be born in clouds of matter containing the production of heavy elements from several stars of population III, stars which could also be found in the form of a multiple stellar system.

It turns out that if this hypothesis is correct, it translates into very specific compositions with abundances of elements which are just as so in the atmospheres of the stars of the EMP populations that can be studied.

Kavli IPMU researchers, led by Tilman Hartwig, then had the idea of using the possibilities offered by machine learning, machine learning as we say in English (a good introduction to AI with machine learning can be found in Jean-Claude Heudin’s book). A supervised machine learning algorithm, fed by theoretical models of supernova nucleosynthesis, allowed them to study and classify more than 450 extremely metal-poor stars observed to date in the Milky Way. They found that about 68% of the EMPs studied have a chemical fingerprint consistent with enrichment by several previous supernovae from population III stars.

In fact, this allows Hartwig to explain in a statement from the Kavli IPMU that “ the multiplicity of early stars was only predicted from numerical simulations, and there was no way to test this theoretical prediction by observations until now. Our result suggests that most early stars formed in small clusters, so several of their supernovae may contribute to the metal enrichment of the early interstellar medium. ».

The same press release also explains that in the near future a much larger number of EMPs and spectral data relating to them will be available, in particular thanks to the Prime Focus Spectrograph equipping the Japanese Subaru telescope in Hawaii.

But the researchers go further. The fact that the first stars could sometimes be found in the form of a binary system in the small clusters that they formed suggests that they could give sources of gravitational waves that the next generations of detectors in space, such as eLisa by the 2030s, will be able to test.

In any case, the study of the first stars can inform us about still mysterious epochs in the history of the observable cosmos, that of the dark ages and especially that of reionization.