Materials design has progressed from studying simple bulk properties to targeting collective effects in strongly correlated materials, such as magnetic ordering, phase transitions, and quantum coherence. This requires a fundamentally different set of computational tools than have been used in the past. The QMCPACK project is employing Quantum Monte Carlo (QMC) methods to attack this problem since they robustly deliver highly accurate calculations of complex materials that do not artificially bias solutions of a given character. By using exascale computing, QMC could attain a 7× increase in the number of modeled atoms or 50× improvement in time to solution to fill missing gaps in experimental data and lead to advances in materials and nanoscience.

Advanced materials design is critical for developing new technologies in fields ranging from energy science and manufacturing to transportation and aerospace. However, engineering lighter and stronger materials or discovering materials with completely new behaviors requires highly accurate models and immensely complex calculations to describe how atomic and molecular systems will interact. Researchers are increasingly turning to high performance computing to understand and accurately predict the quantum mechanical interactions which determine the properties and behavior of a material, taking advantage of new and more powerful machines along with a slew of novel theoretical and computational approaches. Quantum Monte Carlo (QMC) methods are among the most accurate tools for designing accurate materials simulations based on first principles, albeit at a high computational cost. QMC allows for highly accurate calculations and enables feasible testing of any approximations, rendering computational models of this type truer to nature than other methods.

The Exascale Computing Project’s QMCPACK application is designed to apply the power of the world’s fastest supercomputers to meet the enormous computational costs of advanced QMC simulations. Researchers use the QMCPACK software suite to accurately and reliably simulate both novel and known materials with unparalleled scale and efficiency. These simulations reveal material structures and interactions with high confidence and no artificial bias in results, leading to advances in both fundamental research and the design of new materials-based technologies.

QMC provides a solid theoretical basis for accurate and testable materials simulations, but the computational load involved in these simulations has proven prohibitive for all but the simplest of systems, such as materials in their bulk pure form. Without the power of exascale computing, large systems with complex characteristics—such as defects and intentionally added impurities or structural asymmetries—cannot be studied. Because these properties are often key to a substance’s physical properties, exascale-enabled simulations are critical for broadly applying QMC to materials research.

The QMCPACK development team has met the exascale computational challenge by delivering an application that, while running on Frontier and Aurora, is capable of the largest and most accurate QMC simulations to date. The application supports highly accurate simulations of up to around 1000 atoms while accommodating material complexities such as asymmetries and impurities. The QMCPACK team has demonstrated these improved capabilities by simulating a range of scientifically relevant materials such as transition metal oxides—which are commonly used as catalysts and display properties including superconductivity—as well as novel nanoscale materials for optical applications and potential materials candidates for next generation electronic devices.

The advances to materials simulations afforded by QMCPACK will provide important predictive capabilities for materials where previous methods were unreliable. Furthermore, exascale-enabled QMCPACK simulations will allow researchers to perform highly accurate benchmark calculations on a wide range of materials, providing much-needed reference and calibration data for cheaper and less powerful computational methods. These data will be useful across the materials science, nanoscience, and physics communities, particularly where experimental data are costly or difficult to obtain, and can be used directly in machine learning and artificial intelligence applications.

The ability to computationally design, optimize, or understand the properties of energy-relevant materials is fundamentally contingent on the existence of methods that can accurately, efficiently, and reliably simulate them. Quantum mechanics-based approaches must necessarily serve as a foundational role because only these approaches can describe matter in a truly first-principles (i.e., parameter-free) and therefore robust manner. QMC methods are ideal for the required simulations because they robustly deliver highly accurate calculations of complex materials that do not artificially bias solutions of a given character. Significantly, with increased computer power, the few approximations in these methods can be tested and systematically reduced, which is impossible with other first-principles methods.

The trade-off is that the computational demands of the QMC method are large. For example, the use of petascale computers has allowed calculations of the magnetic exchange in a copper oxide, which is important for understanding the mechanism of high-temperature superconductivity. However, these calculations involved a highly symmetric supercell that contains only 56 atoms, whereas a realistic model that considers the defects and dopants of actual superconductors would require at least several hundreds of atoms. The 10 year challenge problem is to simulate transitional metal oxide systems of approximately 1,000 atoms to 10 meV statistical accuracy, such as complex oxide heterostructures that host novel quantum phases, by using the full concurrency of exascale systems. The additional power and parallelism of exascale QMC will provide the essential predictive and quantitative capability for these and related materials that lie well beyond the capabilities of existing methods. Exascale provides the opportunity for highly impactful and enabling benchmark accuracy calculations on these materials, providing the reference calibration data that are missing from essentially all quantum mechanics-based materials calculations today. This capability will be highly useful across the materials sciences, nanoscience, and physics communities, particularly where experimental data are costly or difficult to obtain.

The challenge problem involves calculating the cohesive energy of a large supercell of nickel oxide (NiO) by using QMCPACK and diffusion QMC to an accuracy of 0.010 eV per NiO formula unit at capability scale in a reasonable and scientifically productive amount of wall-clock time (e.g., <1 day). Depending on the memory available on each Exascale machine, the project expects to obtain a solution for a 1,024 atom supercell.

NiO was selected because it is emblematic of science challenges that address the complex physics of transition metal oxides. This classic Mott insulator—more accurately, a charge transfer insulator—defies nonempirical predictions by other methods. NiO is also part of the class of materials being studied by a US Department of Energy Basic Energy Sciences program-funded Computational Materials Sciences Center. Successfully addressing the NiO problem will indicate that a high and productive rate of computational work could be achieved for other challenging materials, including those with strong electronic correlations, novel magnetic states, and novel quantum phases.