Confluence of Numerical Relativity and Physics-Inspired AI for Multi-Messenger Astrophysics Discovery with the Blue Waters Supercomputer

Eliu Huerta Escudero, University of Illinois at Urbana-Champaign

Usage Details

Erik Schnetter, Roland Haas, Philipp Moesta, Maxim Belkin, David Radice, Eliu Huerta Escudero, Antonios Tsokaros, Yufeng Luo, Eamonn O'Shea, Shawn Rosofsky, Sarah Habib, Bing-Jyun Tsao, Asad Khan, Brockton Brendal, Joseph D Adamo, Abhishek Joshi, Nikil Ravi, Sai Sharan Sundar, Pablo Bosch Gomez, Swapnil Shankar, Sebastiaan de Haas, Alexandru Dima

The NCSA Gravity Group pioneered the use of artificial intelligence and high performance computing for Multi-Messenger Astrophysics [1–3]. We demonstrated how to leverage the unique computational resources provided by the Blue Waters supercomputer to initiate and spearhead this disruptive research program, which is now strongly funded by multi-million NSF awards in artificial intelligence and numerical relativity led by PI Huerta and co-PI Haas.

In this proposal we describe research projects to take this innovative research program to the next frontier. Building upon our strong track record producing impactful research with the Blue Waters supercomputer that has been featured on the cover of Nature Reviews Physics [4], and published in leading astrophysics and astronomy journals [2, 5–13], we provide a clear road map of activities that emphasize interdisciplinary and translational research. In particular, we focus on the design of physics-inspired artificial intelligence models to enable the study of complex astrophysical phenomena. Going above and beyond the construction of accelerated artificial intelligence algorithms to detect and characterize gravitational wave sources, we describe how to design forecasting algorithms that predict the merger of binary neutron stars and black hole-neutron star systems tens of seconds before the binary components coalesce. This innovative approach will provide astronomers an early warning system that may optimize electromagnetic and astro-particle follow-up searches, which are critical to study and interpret the physics and astrophysical environments of these sources [14].

The projects we plan to conduct with the Blue Waters supercomputer encompass the construction of the first numerical relativity waveform catalog that describes eccentric, spinning-precessing binary black hole mergers. This waveform catalog will enable us to identify key features that tell apart orbital eccentricity and spin-precession. It is believed that eccentricity drives the dynamical evolution of compact binaries at early-times, while spin corrections tend to affect the binary’s evolution at later times. However, no study has shed light on the coupling of spin-precession and eccentricity at early times, i.e., when the objects are in the process of spiraling into each other. This study will be the f irst of that nature in the literature. Furthermore, we will use these waveforms to explore whether the tell-tale signatures of eccentricity and spin-precession during the inspiral phase may enable the detectability of these systems before they merge. This study is inspired in our recent work [15], where we showed that waveforms that describe quasi-circular binaries, composed of neutron stars and black hole-neutron star systems, may be detected in advanced LIGO noise tens of seconds before they merge. Thus, it is natural to ask whether artificial intelligence may be used to search for patterns in the complex morphology of eccentric and spin-precessing systems at early times to enable their pre-merger detection.

Finally, we will use the numerical relativity catalogs described above to quantify the importance of including higher-order waveform modes for the detectability of eccentric and spin-precessing binary black holes with advanced LIGO. We conducted the first study of this nature in the context of eccentric, non-spinning binary black hole mergers [12]. Our findings in that study established that when we compare the increase in signal-to-noise ratio due to the inclusion of higher-order waveform modes, eccentric systems tend to be significantly louder than quasi-circular systems. We can only guess the exciting physics encoded in higher-order waveforms models of eccentric, spin-precessing mergers, and how we could use them to search for these sources in core-collapse globular clusters and galactic nuclei.

In addition to the research objectives of this proposal, we will continue and expand our student mentoring program. Over the last several years, the PI and co-PI of this proposal have mentored and trained tens of undergraduate and graduate students, who have developed computational skills that have enabled them to secure graduate positions in top ten physics and computer science graduate programs in the US, as well as leadership positions in companies such as Google, Microsoft, NVIDIA, JPMorgan, Goldman Sachs, etc.

Our research team consists of 3 faculty members, 2 senior researchers, 5 graduate students, and 3 undergraduate students. We propose to use our open-source, community infrastructure Einstein2 Toolkit for these simulations and request a total of 0.997 million NH in support for the proposed research.