State of excitement

Customized codes and top-of-the-line supercomputers are enabling teams in three states to enter a quantum world where excited electrons and other particles create and process energy.

“We’re not designing devices, and we’re not making materials,” says Jim Chelikowsky, a University of Texas at Austin professor whose focuses include computational materials, physics, chemical engineering and chemistry. “But we’re helping people who do those things understand how to make them better.”

Chelikowsky is lead principal investigator of a $6.25 million, five-year project under the Department of Energy’s Scientific Discovery through Advanced Computing (SciDAC) program. Its goal is to devise new methods and theories for excited electronic state phenomena in energy-related materials.

These SciDAC researchers can, for example, investigate when a single semiconductor electron interacts with an incoming photon of sunlight in a photovoltaic device, says Jack Deslippe, a Lawrence Berkeley National Laboratory (LBNL) physicist and one of the project’s seven other principal investigators.

Deslippe, lead developer of the Berkeley GW package of codes used in this research, describes how the solar photon first excites the electron above its normal ground, or equilibrium, state to a more energetic one.

The excited electron leaves behind an oppositely charged “hole,” Deslippe says, which itself can quickly merge into “an excited electron-hole pair.” Such energized electrons, holes and electron-hole pairs can behave in special particle-like ways and are thus known as quasiparticles.

Quasiparticles eventually relax back to the ground state and decay. But while decaying, he adds, they can interact with each other strongly and with the rest of the electrons in the system.

‘Excited-state properties are particularly important in describing energy-related devices such as solar cells.’

In a semiconductor, the number of quasiparticles could reach somewhere around Avogadro’s number (roughly 10²³), Deslippe notes – an electron horde too numerous and complex for any analytic mathematical model to handle.

Rather than analytic math, the Berkeley GW code relies on the “many-body perturbation theory,” which he calls a computationally tractable description of how quasiparticles behave in the presence of other electrons. The “G” in the Berkeley package’s name refers to computations based on Green’s functions; the “W” stands for the screened Coulomb interaction, Deslippe says. Both describe quasiparticle behavior in a material. Another important part of Berkeley GW is its formulation of the complex Bethe-Salpeter equation, which can compute how electron-hole pairs get excited and absorb light.

“Excited-state properties are particularly important in describing energy-related devices such as solar cells,” he says. “Learning more about them could lead to new devices.” For example, new types of light-transparent materials could be made that “conduct electricity like a metal. You could eventually paint such conductors on walls to help construct photovoltaic collectors.”

These phenomena, says co-investigator Steven Louie, “are essential to many useful energy generation and conversion processes.” Louie, an LBNL and University of California researcher, led the group that originated the GW codes and introduced an ab initio, or first principles, approach for studying excited-state phenomena.

Ab initio in this case refers to “the theoretical and computational study of material properties from basic quantum theories without any adjustable parameters or empirical input,” says Louie, who also was Ph.D. advisor to Deslippe, a Department of Energy Computational Science Graduate Fellowship alumnus.

Deslippe is quick to note that Berkeley GW software doesn’t act alone. An equally important mathematical modeling code, called PARSEC, was developed by Chelikowsky’s group to describe initial electron behavior at the unexcited ground state.

PARSEC is an acronym, with the “R” and “S” meaning “real space,” the “E” and “C” meaning “energy calculations,” and the “P” standing for “pseudopotentials.” Pseudopotentials “simplify a way of describing electronic states of materials by focusing only on chemically important states,” Chelikowsky says.

Using a method called density functional theory (DFT) and pseudopotentials, PARSEC can simulate the behavior of many more atoms at the ground state than is now possible for the excited state. To accelerate calculations on parallel processing computers, the Texas team also invents algorithms that divide the task of computing an immense number of different electronic states.

“If I have a material where none of the electrons are excited, what properties would it have, and how are the electrons distributed in energy and space?” Chelikowsky asks. “With such a ground state problem, it’s easy – relatively speaking – to do systems of up to 20,000 atoms.” But modeling the excited state “is a much harder problem. At this time, if I could do (excited) systems with a few thousand atoms I’d be pretty happy.”

PARSEC is “really an impressive code” that “Berkeley GW can sit on top of,” Deslippe says. “We are coupling the best-in-class DFT code with the best-in-class GW code to study the ground state and the excited-state properties of large systems of interest.”

He and other SciDAC group members load those complementary codes onto top supercomputers such as Edison, a Cray XC30 supercomputer with a peak performance of 2.39 petaflops (quadrillion calculations per second) at LBNL’s National Energy Research Scientific Computing Center, or Mira, a 10 petaflops IBM Blue Gene/Q at the Argonne Leadership Computing Facility at Argonne National Laboratory.

Meanwhile, Jeff Neaton, another project principal investigator, heads LBNL’s Molecular Foundry, a national user facility supported by the Department of Energy Office of Basic Energy Sciences through their Nanoscale Science Research Center program. There, visiting scientists conduct fundamental research at the nanoscale to develop new materials and to understand them through new instrumentation and computational methods.

With focuses that include artificial photosynthesis and energy storage, the Molecular Foundry provides experimental backup for Berkeley GW and PARSEC. Deslippe notes that its researchers “drive a large number of applications that in turn drive the development of the codes.”

Neaton, also is acting director of LBNL’s materials science division, joined Louie and three others to use the ab initio method to compute, for the first time, how fast heated electrons and holes, or “hot carriers,” lose their excess energy in a semiconductor. Hot carriers can make a solar cell less efficient because much of the sunlight they absorb is shed via phonons – units of quantum crystal vibrations – instead of generating electricity.

The hot carriers finding is just one more example of where a SciDAC collaboration can lead.