Computing Sciences masthead Berkeley Lab Computing Sciences Berkeley Lab logo

Exascale for Energy

Scaling-up Computational Methods for Solar Energy Research

Multicore and heterogeneous computer architectures are setting new records for computing speed and are demonstrating possible paths toward exascale computing. But taking full advantage of faster hardware for solar energy and other fields of research requires new and improved mathematical methods and computational tools.

Researchers in Berkeley Lab’s Computational Research Division (CRD) are collaborating on three projects that are working to improve algorithms and codes specifically for solar energy research; but the methods and tools they develop will also have applications in combustion efficiency, materials science, and many other scientific domains.

The computational methods and tools developed for solar energy research will also have applications in combustion efficiency, materials science, and many other scientific domains.

Computational chemistry codes such as GAMESS, NWChem, and MPQC are among the most widely used in the DOE research community and beyond, with applications in solar energy cell design, combustion efficiency, materials science, nanoscience, nanoelectronics, and related fields. Tuning these codes by hand to make them run efficiently on a particular computer usually requires a high level of expertise and lots of time, especially as computers grow in size and complexity.

One promising way to speed up the process is autotuning—the development of tools and techniques that can automatically generate and test variations of a scientific code, resulting in a tuned code that achieves very high performance on a given system.

The “Autotuning Large Computational Chemistry Codes” project is implementing state-of-the-art performance analysis and autotuning techniques to accelerate some key computational chemistry applications, notably a linear scaling multi-reference configuration interaction (MRCI) module in the GAMESS code. One particular application targeted is a set of large-scale simulations of large hydrocarbons and sulfur-containing hydrocarbons that are components of diesel fuel.

Collaborators include two DOE Energy Frontier Research Centers (EFRCs)—the Combustion EFRC (CEFRC) and the Argonne-Northwestern Solar Energy Research Center (ANSER)—as well as researchers from Lawrence Berkeley, Ames, and Sandia national laboratories and the universities of Tennessee and Oregon.

The “Large-Scale Eigenvalue Calculations in the Study of Electron Excitation for Photovoltaic Materials” project is a collaboration between the SciDAC TOPS (Towards Optimal Petascale Simulations) center and two EFRCs: the Center for Inverse Design at the National Renewable Energy Laboratory (NREL), and the Molecularly Assembled Material Architectures for Solar Energy Production, Storage, and Carbon Capture EFRC at the University of California, Los Angeles (UCLA).

This project is focusing on developing and deploying state-of-the-art solvers for specific large-scale eigenvalue problems. The improved algorithms will help researchers search for nanostructure-based effects for the design of new photovoltaic materials with much higher efficiency, as well as improve the efficiency of organic solar cells and design new solar cell types. The project aims to improve code scalability and performance on modern multicore supercomputers.

Production of solar energy can take many forms, from the direct production of fuels from sunlight through artificial photosynthesis, to the production of electricity through photovoltaics. Using computation to identify promising classes of chemicals that can catalyze such reactions involves multiscale QM/MM (quantum mechanics/molecular mechanics) simulations coupled with higher-level models, all under the coordination of an optimization framework to search for candidate molecules.

The project “Enhancing Productivity of Materials Discovery Computations for Solar Fuels and Next Generation Photovoltaics” brings together a group of computer scientists and applied mathematicians from the SciDAC Performance Engineering Research Institute (PERI) and the Solar Fuels and Next Generation Photovoltaics EFRC at the University of North Carolina. The three major thrusts of this effort are: (1) integrating performance engineering tools and methods with the ongoing development of the chemistry codes; (2) developing advanced simulation-driven optimization methods to apply to this problem; and (3) structuring the computations under a workflow and data management framework to ensure performance portability and productivity across a range of computing systems.


<< Previous page