UC scientists use machine learning to reduce neutron star calculation time sevenfold

Neutron stars are among the densest celestial objects in the universe. They can be viewed as giant, neutron-rich nuclei, but their true composition remains a mystery. Are quarks deconfined within them? Could their structure include hyperons - particles similar to nucleons, but containing a strange quark, besides protons and neutrons?

Using symbolic regression, a machine learning technique that identifies algebraic relationships between different properties, a research team from the Department of Physics of the University of Coimbra's Faculty of Sciences and Technology (FCTUC) uncovered a correlation between the maximum mass of a neutron star and its equation of state.

This discovery, published in Physics Letters B, reduces the time required for one of the most critical steps in identifying models compatible with astronomical observations by a factor of seven. This step relies on Bayesian inference, which can be highly time-consuming due to the need to solve differential equations that describe the mass and radius of several million star models.

Constança Providência, a researcher at the University of Coimbra’s Centre for Physics (CFisUC) and professor at FCTUC, explains that current experimental and observational data may eventually help us understand what these objects are made of. However, she adds that it is still very hard to study matter held together by the strong nuclear force—the force that keeps protons and neutrons tightly bound in the centre of atoms. These particles are made up of even smaller ones called quarks. This kind of matter, especially when found at extremely high densities, like in the cores of neutron stars, is still very hard to investigate.

Using statistical methods has been crucial in advancing this line of research. However, the researchers state that identifying the equation of state of matter under the extreme densities and pressures inside neutron stars based on their known masses and radii is a complex task. This is because a large number of models must be tested, which requires significant computing time.

"We hope that we will soon be able to decode the equation of state of dense matter directly from precise observational data on neutron stars, using advanced computational techniques. This will enable us to determine the properties of baryonic matter at high densities and establish the density at which quarks become deconfined from nucleons. It will also help us establish whether the transition to deconfined matter is a first-order phase transition," they conclude.

The scientific article “Inferring the equation of state from neutron star observables via machine learning” is a collaborative effort between researchers Tuhin Malik, Helena Pais, and Constança Providência from CFisUC, alongside scientists from China and India. It is available here.