Using the KKR-CPA method, the MuST software converts the complex surroundings of an atom in a random alloy (brass, an alloy of copper and zinc, in this case) to an “effective medium” that averages the properties of the surrounding atoms.

Why It’s Important

Materials science is important for making our world run better, and at less expense. Doping a silicon wafer can change it from being an electrical conductor to an insulator to a semiconductor—one of those miracle substances we take for granted but which make our computer-aided modern lives possible. Altering the ratio of atoms in a glass mixture can produce a nearly unbreakable smartphone screen. Changing the composition of a metal can make it stronger, lighter or easier to manufacture and form to shape.

The problem is, materials science is complicated. Particularly when a material has many different elements in it, understanding its properties—and how it can be engineered to do what we want—involves taking into consideration the interactions between each atom, its neighbors, their neighbors, and so on. This many-body problem is technically impossible to solve exactly. But computers can give us an approximate solution to a high level of confidence. With the complicated rules of quantum chemistry that govern materials at the atomic scale, though, the complexity ramps up when you try to simulate more than a few atoms of a material. This is a problem, as small numbers of atoms may have different properties than the same material in real-world bulk. Particularly for disordered substances—some of the most promising and interesting materials for development—this complexity quickly pushes the problem beyond the point at which even the world’s largest supercomputers can crunch it.

How PSC Helped

Enter Yang Wang, senior computational scientist at PSC, and an international collaboration with the Oak Ridge National Laboratory, Universität Augsburg, University of the Chinese Academy of Sciences, Louisiana State University and Middle Tennessee State University. Through the NSF’s Cyberinfrastructure for Sustained Scientific Innovation program and XSEDE ecosystem of supercomputing resources’ Extended Collaborative Support Service, they’ve developed MuST, a software package that uses density function theory (DF) for ab initio investigation of disordered materials—that is, predicting materials’ properties from first principles.

Ab initio quantum chemistry method is a time-tested way to accurately predict the properties of a substance. It also helps lab scientists focus expensive real-world experiments on the most promising candidates, speeding development. But DFT calculations typically scale with the third power of the number of atoms—in other words, if computing the behavior of a given number of atoms in a material takes a certain amount of time and power to calculate, twice as many atoms will take eight times as much, three will take 27 times as much and four times as many atoms 64 times as much. This puts a tight limit on how many atoms can be simulated.

MuST takes advantage of a combination of the Korringa-Kohn-Rostoker and coherent potential approximation methods (KKR-CPA) to simplify the problem. Instead of, say, calculating the interactions of an aluminum atom in an Al-Cr-Fe-Co-Ni alloy with each aluminum, chromium, iron, cobalt, and nickel atom nearby, it calculates those other atoms as a kind of average “soup” in which the aluminum atom sits.

MuST also takes advantage of the locally self-consistent multiple scattering theory (LSMS) to enable large supercell calculations for the study of disordered structures. The LSMS method reduces the complexity of the computation for supercells enormously. Instead of scaling to the third power, it scales with the number of atoms. In other words, instead of 64 times the computing power enabling you to calculate four times as many atoms in the supercell, it enables you to simulate 64 times as many. Initial work with MuST, which has been available to the general scientific community since December 2019, has produced results as good as or better than those of gold-standard methods that require much more computing power. It also goes beyond the reach of those methods when large numbers of atoms (thousands or more) are involved. More recently, the National Science Foundation has approved a proposal by the group to give MuST the capability to calculate the electrical and spin conductivity of a material, which will further flesh out the detailed behavior of materials in the program.

One goal for the future is incorporating the LSMS method with typical medium embedding to allow for capturing the metal-insulator transition phenomena driven by disorder in quantum materials. Another is integrating the Kubo-Greenwood formula into the package, a project in which Wang has advised Vishnu Raghuraman, a graduate student in Mike Widom’s group at CMU. This work, which Raghuraman is currently preparing as a peer-reviwed journal paper, will enable the investigation of electronic transport in disordered structures—the movement of electrons through a material. This phenomenon affects a material’s properties and underlies important chemical processes like photosynthesis. Scientists can download the program—and join the team to develop MuST—here.