They upgraded the BerkeleyGW open-source software to support dynamic, many-body simulations that couple electron motion with nuclear motion, moving beyond static pictures of electron behavior.

The effort builds on more than a decade of BerkeleyGW development, optimizing performance across Intel, AMD, and NVIDIA GPUs to ensure portability across evolving hardware architectures.

Aurora’s large memory and scalable architecture enabled memory-intensive simulations of systems with tens of thousands of atoms, enabling dynamic simulations that were previously unattainable.

The project, titled “Advancing Quantum Many-Body GW Calculations on Exascale Supercomputing Platforms,” involves USC, Berkeley Lab, Stanford, Oak Ridge, and NREL, with access through DOE’s INCITE program.

Researchers from USC and Lawrence Berkeley National Laboratory advanced simulations of quantum materials using three DOE supercomputers—Aurora, Frontier, and Perlmutter—to achieve greater precision in their models.

Simulations reached over one exaflop on Frontier and more than 0.7 exaflops on Aurora, setting new benchmarks for exascale quantum-mechanical calculations.

Using GW perturbation theory within a single framework, the team models electron-phonon interactions to make predictions crucial for nanodevice design and understanding phenomena like superconductivity and transistor performance.

BerkeleyGW 4.0 has been released to the broader research community, expanding access to these advanced capabilities.

The study highlights potential impacts on quantum materials research and applications in computing, energy, and electronics, including improved band-gap predictions and materials design from theory.