“Where did we come from?”
That is the complex question that Columbia University astronomy professor Greg L. Bryan works to answer as director of a new Simons Collaboration to learn the universe. His goal: To go beyond our current, unsatisfactory understanding of the universe and its contents, and to leverage emerging technologies and methodologies to better understand the initial conditions of the universe and the physical laws that have governed its evolution.
“We’re looking for the initial distribution of matter and the laws of physics that got us from there to here over the 13.7 billion years that the observable universe has existed,” Bryan, who is also a Data Science Institute affiliate, explained. “If we can map out to the edge of what’s known, then hopefully that can tell us what lies beyond that edge.”
The Learning the Universe collaboration is ambitious, multidimensional, and multidisciplinary. It brings together experts from across disciplines and universities from the fields of cosmology, galaxy formation, machine learning, and statistical inference. It will use cosmological observation, next-level simulations, and machine learning to model galaxy formation and infer the initial conditions and cosmological parameters of the universe. The final result will be a new cosmic map with a better understanding of our place in it.
Astronomers expect an exponential increase in data over the next few years as new observatories are mapping more and more of the universe to better understand the universe’s key components—dark energy and dark matter. “The amount of data coming down the pipe towards us is mind-boggling,” Bryan said. “It is one of the motivations behind this project—to build a framework to use and understand the data when it gets here.”
At the same time, computational capabilities have increased to a point where innovative approaches to model building and galaxy formation simulations may enable scientists to leverage the increase in data to answer some of the toughest and oldest questions.
In order to understand the initial conditions and the forces giving rise to expansion and acceleration, Bryan and his colleagues will need to work backwards. “We’re trying to ‘go back in time’ to understand where we came from. What is dark energy? How is it driving acceleration? Mapping the trajectory can help us infer something about the forces that are acting on it,” he said.
Learning the Universe involves three interconnected components: improving cosmological simulations, including galaxy formation simulations; accelerating the process of forward modeling by using these simulations as training data for machine learning algorithms; and using these rapid models along with the observations to infer cosmological parameters and initial conditions.
Bryan and his team focus on a small patch of the universe, take a guess for its initial conditions, develop models that propagate forward from this initial estimate, and compare the computational predictions with the real universe. By repeating this process many times, the collaborators are able to determine which initial conditions produce better matches. Successful models are used to train machine learning algorithms to work with more and more observational data. “Each forward modeling computation is very energy and resource intensive and needs to be done many times. It requires a lot of guesses and a lot of computation,” Bryan noted.
The Simons Foundation‘s four-year grant will support the collaboration between eight institutions and principal investigators under Bryan’s direction. Experts in galaxy formation create simulations and predictions, while machine learning experts create models that can learn and rapidly reproduce results. Statistical inference experts infer the initial conditions and key parameters, and experts in cosmology provide context for the collaboration.
“It’s been a huge joy to put this together and find the right people willing to take this leap and join us in this project. I’ve also had to step outside my comfortable boundaries and talk to people about things I know very little about,” Bryan admits. “Clarity around this project and how we describe the work is one of the advantages of having to collaborate across disciplines. You’re forced to explain your work to people outside your field.”
— Karina Alexanyan, Ph.D.