Machine learning and neutron star physics


  • Call:

    IDPASC Portugal - PHD Programme 2019

  • Academic Year:

    2019 / 2020

  • Domain:


  • Supervisor:

    Constança Providência

  • Co-Supervisor:

    Márcio Ferreira

  • Institution:

    Universidade de Coimbra

  • Host Institution:

    Universidade de Coimbra

  • Abstract:

    There is a growing interest in applying Machine Learning (ML) techniques in astrophysical problems. The challenging issues faced in astrophysics, and in particular Gravitational Waves (GW) physics, demand for different perspectives and approaches of computer science. The efficiently handling of complex and massive data sets demands for new techniques and algorithms. One of the open questions in nuclear physics is determining the equation of state of nuclear matter. Neutrons stars (NSs) are singular physical systems that allow one to investigate the equation of state of nuclear matter under extreme conditions, far beyond the ones attainable in laboratory experiments. The GWs signals emitted during a neutron stars merger carry crucial information about the neutron star matter at high densities. The tidal deformability is one important signature carried by GWs that constrain the equation of state of nuclear mater. Advanced LIGO and Advanced Virgo gravitational-wave detectors made their first observation of a binary NS inspiral on August 2017. The GW170817 observation exposed the potential to directly probe the physics of NSs, which opened a new era in the field of multi-messenger astronomy and nuclear physics. From the analysis of the GW170817 data, was possible to set an upper bound on the NSs combined dimensionless tidal deformability. Further insights into the physics of neutron stars are expected from future gravitational wave observations. Bayesian analysis is one of the standard methods for making inference of physics information from experimental observations. The Bayesian inference relies in calculating the posterior probability distribution, which is the product between the likelihood function and the prior probability distribution. The prior distribution encodes our present knowledge on some physical quantity and, for a limited number of available data, its choice becomes crucial for the Bayesian inference. Therefore, only for a sufficiently larger data-set, the inference becomes weakly dependent on the prior. Machine learning methods might be a reliable alternative for inference even when we are faced with a limited experimental data-set. Deep learning is a branch of machine learning with a hierarchical structure of neural networks that extracts high-level representation from data [1]. Due to the non-linear representations that these hierarchical structures can learn from data, they are highly effective in tackling complex non-linear systems. They have been very successfully applied in images or speech pattern recognition and automated translation. Currently, there is an increasing interest in the application of these Deep learning methods in many areas of physics. In condensed matter physics, they were applied in the identification of phase transitions [2]. In particle physics, for the processing of experimental heavy-ion collision data-set [3]. Recently, they were employed for recognizing the different phases of a quantum field theory system and to predict the value of several observables [4]. The use of Deep learning on neutron star physics was first explored in [5]. A neural network was employed as an efficient procedure for mapping from a finite set of mass-radius data with observational errors onto an equation of state. The presents project proposes to further explore these Deep learning methods in constraining the equation of state of nuclear matter. It possibilities physical inference on the equation of state properties from the combined experimental and observational results. The generation of the data-set on which the Deep learning methods will learn is a crucial step, which must consist of all physical reliable equations of state. [1] Mehta, Pankaj, et al. "A high-bias, low-variance introduction to machine learning for physicists." Physics Reports (2019). [2] Carrasquilla, J., \& Melko, R. G. (2017). Machine learning phases of matter. Nature Physics, 13(5), 431. [3] Pang, Long-Gang, et al. "An EoS-meter of QCD transition from deep learning." , Nature Commun. 9 (2018) no.1, 210\\\\ [4] Zhou, K., Endrődi, G., \& Pang, L. G. (2018). Regressive and generative neural networks for scalar field theory. arXiv preprint arXiv:1810.12879.\\\\ [5] Fujimoto, Y., Fukushima, K., \& Murase, K. (2018). Methodology study of machine learning for the neutron star equation of state. Physical Review D, 98(2), 023019.