Five Michigan State University (MSU) researchers from the Facility for Rare Isotope Beams (FRIB) (frib.msu.edu) and the Department of Statistics and Probability (STT) (stt.natsci.msu.edu) are participating in a new $3.7 million National Science Foundation (NSF) project to advance nuclear physics experiments.
The multi-institutional effort will develop software to create more accurate models of scientific phenomena--from what happened in the microseconds after the Big Bang to how long a radioactive nucleus will live before it decays.
The project will bring together statisticians, computer scientists, and nuclear physicists from MSU, Ohio University, The Ohio State University, and Northwestern University. They will develop a framework that uses Bayesian statistics to increase modeling accuracy. It will combine features from several models into a more powerful and predictive one.
The Bayesian Analysis of Nuclear Dynamics (BAND) framework will provide a publicly available set of computational tools for physicists seeking to solve a wide variety of nuclear-physics research questions.
The MSU team includes Frederi Viens, chair and professor in STT, who is also the director of Actuarial Science and adjunct director of Statistical Training and Consulting. Viens works on applications of Bayesian statistics in many disciplines. He will lead coordinating the work at MSU with the other institutions.
The four other MSU team members are Tapabrata Maiti, MSU Foundation Professor, also in STT, a specialist in developing Bayesian methodology; and three prominent FRIB theorists who apply Bayesian who apply Bayesian methodology to nuclear modeling: Witek Nazarewicz, John A. Hannah Distinguished Professor of Physics and chief scientist at FRIB; Filomena Nunes, managing director for the FRIB Theory Alliance and professor of physics; and Scott Pratt, professor of physics.
In a classical use of the scientific method, scientific theory is updated based on observations from experiments. Experiments are prompted by efforts to validate the theory. The Bayesian context enriches this discovery cycle by providing systematic ways to analyze data and uncertainty. Statisticians and domain specialists work hand in hand to translate each other's ideas and beliefs into sound statistical models.
"The basic question that BAND will address is: What is the best way to use experimental data in the formulation of theoretical models that attempt to explain the results of experiments and make predictions for new observables, often involving huge extrapolations-" explained Nazarewicz.
Daniel Phillips, a professor of physics and astronomy at Ohio University, leads the effort. Phillips has worked with physicists on experimental design for the 20 years he's been a member of Ohio University's Institute for Nuclear and Particle Physics. He has served as director of the institute since 2014. Five other scholars from Ohio State and Northwestern complete the team's coverage in nuclear physics, computer science, applied mathematics, industrial engineering, and statistics.
Because current models can yield very different forecasts, scientists hope the project will improve the characterization and reduction of uncertainties, or uncertainty quantification, for a range of nuclear processes. BAND's predictions for those processes will be phrased in terms of the probability of different things happening--somewhat like the National Weather Service's rain forecast, Phillips explained.
"One of the goals is for BAND is to be accessible to people planning nuclear physics experiments and allow them to design in such a way that cuts down the uncertainty as much as possible," Phillips said. "The U.S. invests hundreds of millions of dollars in nuclear physics experiments--optimal allocation of those resources is important."
The five-year grant from NSF's Cyberinfrastructure for Sustained Scientific Innovation program will support regular software releases, with functionality growing annually. The award will also fund workshops to train scientists on how to use the computational tools.
The four universities have been developing the concept for BAND since 2016, sparked during a think tank about how nuclear physicists can incorporate Bayesian analysis in their work. Viens and other MSU team members started in the Bayesian direction for nuclear physics at that time. The BAND framework grant will enhance the collaboration between nuclear physics and the statistical sciences at MSU (frib.msu.edu/news/2017/statistical-science-researcher.html) that was launched in 2017.
"This framework is particularly well suited to developing a principled honest and efficient uncertainty quantification, and has features that enable model mixing in a natural way, even when models cover domains that only overlap partially.
This makes it very attractive for nuclear physicists," Viens said. "They also appreciate being able to interpret every result as the probability of some event of interest; and they like the machine-learning features of some of the computational algorithms we use. I find the application to nuclear physics fascinating because nuclear physicists are not afraid of computational challenges, and love mathematical modeling."
"The BAND framework is about speeding up the cycle of the scientific method through advanced statistical tools," said Nazarewicz. "These new tools will provide meaningful input for planned measurements at FRIB and will be used to interpret and use new nuclear and astrophysical information obtained at FRIB.