A team working on biofuel research is rewriting the decades-old NWChem software program for the exascale era. The new software, NWChemEx, will enable computational chemistry software to effectively run on exascale supercomputers. The team, part of Argonne National Laboratory’s Early Science Program (ESP), estimates that the future Argonne Aurora supercomputer, along with the new NWChemEx algorithms, will provide up to a 1,000x increase in the size of the chemical systems that can be studied.
Distinguished Professor Theresa Windus, Ames Laboratory and Iowa State University, is the principal investigator on the project. Windus states, “One of the goals of our research is to find more accurate information about chemical systems to aid in developing new methods for converting biomass into biofuels. When the Aurora supercomputer comes online at Argonne, it will be one of the largest supercomputers available. Running computer simulations of our biofuel research on Aurora using the NWChemEx software will allow the team to investigate larger molecular systems and perform deeper exploration of the energy landscape for molecules and materials. That additional knowledge provides a better understanding of the fundamental chemical reactions and their properties.”
In addition to Windus, the research team involves a broad spectrum of expertise located at Ames Laboratory, Argonne National Laboratory, Brookhaven National Laboratory, Lawrence Berkeley National Laboratory, Oak Ridge National Laboratory, Pacific Northwest National Laboratory, and Virginia Tech, and frequent collaborations with Intel, AMD, HPE and Nvidia.
Updating NWChem code for Exascale computing
NWChem is a widely used open source computational chemistry package that includes both quantum chemical and molecular dynamics functionality. However, the NWChem code was designed for computer architectures that no longer exist. NWChemEx is a complete rewrite of the software – it’s modularized and flexible from its initial design and uses the power of C++ – enabling the team to move away from the megalithic design pattern that’s plagued quantum chemistry since the 70s. A primary focus of NWChemEx is designing the code so it runs on graphic processing units (GPUs) and uses dynamic methods to shift computational work to optimize performance.
“This change lets us develop methods to optimize the code over time and not force the code to run on a specific implementation. In terms of optimization, we are enabling different levels of parallelization within and across tensor contraction expressions. This allows us to tune the parallelization strategy to the specific execution,” states Sriram Krishnamoorthy, a laboratory fellow and computer scientist at Pacific Northwest National Laboratory. The change will also enable NWChemEx to run on the Intel Ponte Vecchio GPU units that will be incorporated in Aurora.
The key points are modernization of the Density Function Theory (DFT) code that is widely used in quantum chemistry, and the Coupled Cluster (CC) theory that is used to access the high accuracy required for predictive chemistry. The team is focused on implementing reduced scaling chemistry algorithms that challenge implementation on high performance, heterogeneous hardware due to nonuniform sparsity in these methods.
David William-Young, a research scientist at Lawrence Berkeley National Laboratory, states, “To aid in the implementation, the team uses modularization of the algorithmic components (grid/batch generation, load balancing, local work drivers, reduction, etc.), so they can be swapped and sandboxed individually to rapidly test new ideas and architecture-specific implementations without interfering with other components. The module approach opens the door for reuse of software written for one purpose in other parts of NWChemEx.”
Resolving computational issues in NWChemEx
“New exascale hardware requires a new software environment. The team is modifying the code to keep up with all the changes in hardware architecture by adapting to a new and unifying programming model – oneAPI. The knowledge gained from the NWChemEx project can be used and shared when Aurora is operational,” states Abhishek Bagusetty, Aurora Early Science Program Postdoctoral Appointee, Argonne Leadership Computing Facility (ALCF).
ALCF is a U.S. Department of Energy (DOE) Office of Science User Facility located at Argonne National Laboratory. The ALCF team works with researchers to help enable breakthroughs in science and engineering by providing supercomputing resources and expertise to the research community.
“The team is working to resolve issues such as load-balancing occurring on the Oak Ridge National Laboratory Summit supercomputer in preparation for Aurora. Methods used to resolve communication and load balancing will apply outside of the NWChemEx code,” indicates Ajay Panyala, computer scientist at Pacific Northwest National Laboratory.
Using NWChemEx to find catalytic materials for conversion of biomass into biofuels
The development of advanced biofuels is driven by both energy security and climate change considerations. The Department of Energy (DOE) has an advanced biofuels program to develop fuels that can use the existing infrastructure and replace existing fuels with biofuels on a gallon-for-gallon basis. Windus states, “Producing high-quality biofuels in a sustainable and economically competitive way is technically challenging, especially in a changing global climate. The NWChemEx project directly addresses one of DOE’s priority goals to develop high-performance computational models demonstrating that biomass can be a viable, sustainable feedstock for the production of biofuels and other bioproducts. Of particular interest in this project is the prediction of specific, selective and low-temperature catalytic conversion of biomass to fuels and other products.”
Zeolites, such as H-ZSM-5, offer great promise for the catalytic conversion of renewable biomass-derived alcohols into fuels and chemicals. Compared to metal oxides with diverse surface and acid properties, acidic zeolites are solid structures that have relatively well-defined and uniform acid site structures. This makes it easy to perform rigorous kinetic and theoretical investigations of the effect of acid strength and solvation environment and confinement on the chemical reaction energies. Although there have been a number of prior atomic-scale computational studies of these systems, unraveling the true complexity of the conversion process and identifying means of achieving conversions at lower temperatures and pressures is an unsolved problem.
To illustrate the capability of NWChemEx for chemical reactions, the project will examine a number of elementary chemical transformations that have been postulated for the conversion of propanol to propene in the H-ZSM-5 zeolite (basic unit cell: Si96O192). The team will run calculations using computer simulations to predict the binding energy of water and propanol and their reactions in the zeolite cavity to help identify appropriate biofuels.
“By running the millions of simulations on Aurora, and then identifying a smaller subset of possible chemical solutions, the task of conducting physical experiments will be facilitated and therefore speed the discovery of real solutions,” states Windus.
Intel solutions used in developing NWChemEx
NWChemEx uses Data Parallel C++ (DPC++) as one of its programming models. DPC++ is a cross-architecture programming language, based on C++ and SYCL and part of oneAPI – an industry initiative to unify and simplify application development across diverse computing architectures. To facilitate porting of any existing CUDA code, Intel’s DPC++ Compatibility Tool helps migrate that code to create new DPC++ code. Further performance analysis and tuning uses Intel VTune Profiler and Intel Advisor. The compilers (C, C++, Fortran), Intel oneAPI Math Kernel Library (oneMKL) and Intel oneAPI DPC++ Library (oneDPL) – all part of Intel’s oneAPI Base Toolkit, are also employed.
Future benefits of NWChemEx and exascale supercomputers
The impact of the team’s biofuel research and rewriting NWChemEx software will have benefits outside the Argonne ESP project. “Chemistry simulations need to be able to model more complex, realistic systems. With current hardware and software, we can only simulate simplistic systems by removing impurities, getting rid of defects, and simplifying the solvent environments. In the future, researchers must be able to do real-life simulations that bring in all the complexities of the chemical system to have confidence the research results show the real-life issues, and the research findings will have comparable results to experimental conditions,” states Windus.
“Quantum chemistry research methods have become ubiquitous in hard science and engineering and are used in a variety of fields. What makes molecular dynamics and quantum chemistry tools such as NWChemEx so valuable is that they provide predictive results that are accurate when compared with many other research algorithms. Tools such as NWChemEx and access to hardware such as the future Argonne Aurora supercomputer will be critical to finding future solutions,” states Eric Bylaska, chemist at Pacific Northwest National Laboratory.
NWChemEx research was supported by the Exascale Computing Project, a collaborative effort of the U.S. Department of Energy Office of Science and the National Nuclear Security Administration.
Linda Barney is the founder and owner of Barney and Associates, a technical/marketing writing, training, and web design firm in Beaverton, Oregon.