Mapping cosmic shear to illuminate dark energy

Gravitational lensing often evokes images of a cosmic funhouse mirror: duplicated galaxies, dramatic arcs and distorted shapes. But the web-like, large-scale structure throughout the universe also bends light in a weaker, less obvious way. This phenomenon, known as cosmic shear, can provide clues about the role of dark energy in shaping the universe.

In a recent study published in the Astrophysical Journal, researchers from Lawrence Livermore National Laboratory (LLNL) developed an innovative approach to map cosmic shear using linear algebra, statistics and high-performance computing. With this model, they transformed simulated shear data from specific points into predictions of shear across the sky, effectively filling in observational gaps. The method can handle datasets about 1,000 times larger than previous approaches.

The team focused on quantifying convergence, a measure of how much mass is responsible for lensing at a given location.

“Essentially, our maps create a visual representation of convergence at different points in the window of the sky in which we are looking,” said LLNL scientist and author Greg Sallaberry. “If we build these convergence maps at different points in cosmic time [different distances from us], we can start to piece together a history of how structure has evolved across the universe and uncover the role that dark energy plays.”

However, creating these maps becomes increasingly computationally challenging as the volume of data grows. With the advent of next-generation wide-field surveys, such as those from the Vera C. Rubin Observatory, researchers will need a scalable method to handle the unprecedented influx of data.

To address this, the team optimized their model by focusing only on nearby data points. Each measurement of convergence was treated as being influenced primarily by its closest neighbors, rather than being tied to every other location in the sky.

“The computing environment and software stack that makes implementing our model at scale on high-performance computing systems possible was a decade in the making,” said LLNL scientist and author Min Priest. “It involves many general-purpose software libraries developed at the Lab for large-scale problems.”

The study used simplified simulation data, which may not fully reflect the complexities of real-world astronomical surveys. The authors aim to make it more generalizable going forward.

“The endgame is that we want to have a method and associated product that can work out-of-the-box to generate shear maps in a realistic environment,” Sallaberry said.

Other LLNL authors include Robert Armstrong, Michael Schneider, Trevor Steil and Keita Iwabuchi. Amanda Muyskens, formerly of LLNL, was chiefly responsible for designing the scalable Gaussian process model. Funding was provided by the Laboratory Directed Research and Development program.