National Center for Computational Sciences

Researchers hit jackpot with CASINO code

The image shows how electrons rearrange when four water molecules bond to a sheet of graphene. Image courtesy Angelos Michaelides, University College London.

Some research has potential that’s difficult to quantify. The ramifications of a better understanding of a particular phenomenon can sometimes reverberate throughout science, making everlasting waves in multiple arenas.

Take the seemingly trivial properties of a water molecule on graphene, or a single layer of graphite. Water is everywhere, as is carbon. Big deal, right? Actually, it is. Understanding the adsorption, or accumulation, of a water molecule on graphene could influence everything from hydrogen storage to astrophysics to corrosion to geology.

In 2009 a team made up of Dario Alfe, Angelos Michaelides, and Mike Gillan of University College London (UCL) used the Oak Ridge Leadership Computing Facility’s (OLCF’s) early petascale science period to obtain a full-binding energy curve between water and graphene, research that will impact multiple domains and could only be performed on the world’s fastest supercomputer. The project is a collaboration between Oak Ridge National Laboratory (ORNL) and UCL and is supported by the Engineering and Physical Sciences Research Council of the United Kingdom.

Adsorption is the accumulation of atoms or molecules on the surface of a material. The process is possible because unlike the atoms in a material that bond with one another, surface atoms are not wholly surrounded by other atoms and can attract adsorbates, in this case a water molecule.

While the process may sound a bit strange, it turns out that this is the case with most natural surfaces—they are covered by a film of water. It is for precisely this reason that this research is so fundamental. “Water interacts with every surface that we see,” said Alfe.

Because of the large numbers of particles and multiple variables involved in the interaction, these systems are difficult to describe. However, science could greatly benefit from a more precise explanation of such systems simply because they occur so often in the natural world.

And as science benefits, so do we. For instance, a better understanding of this particular adsorption process could pave the way for more efficient hydrogen batteries, an integral part of a potential clean energy economy; catalysis, the process of adding an agent to initiate a chemical change, is an important step in countless industrial processes—it could be streamlined to make products safer and cheaper; and adsorption is key to understanding lubrication to reduce friction between moving parts, which plays a role in nearly every nook and cranny of our transportation sector.

And that’s just energy. The research will likewise shed light on some of science’s most fundamental questions, such as how the Earth forms materials at the extreme conditions deep in its interior, or how molecules form in outer space’s interstellar medium. And the methodological advances that result will lead to better predictions of the interaction of water with other materials such as clay particles and proteins, both of which are central to biological and environmental research programs.

Skinning Schrödinger’s Cat

Alfe and his team received 2 million hours on Jaguar to calculate a water molecule’s adsorption energy and geometry on a sheet of graphene. Using the quantum Monte-Carlo (QMC)–based CASINO code developed at Cambridge University, the team produced an unprecedented depiction of water-graphene adsorption.

By calculating the adsorption energy and geometry in 30 individual steps as the molecule moves closer to the surface, the team made history. “We’ve been able to obtain a full-binding energy curve between water and graphene,” said Alfe. This breakthrough was only possible using CASINO’s QMC-based methodology.

Using QMC methodology scientists attempt to solve the quantum mechanical Schrödinger equation, which describes the evolution of a quantum state over time, by simulating a mathematically equivalent random process, such as flipping a coin over and over to predict the chance of achieving heads. The more times one flips the coin, the more precise the prediction.

While there are several ways to solve the equation, QMC delivers the most accurate representation. It is computationally intensive, though, requiring the generation of random numbers across tens of thousands of compute nodes—a technique 1,000 to 10,000 times more expensive, but 10 times more accurate—than standard methods such as density functional theory (DFT).

These standard tools are “okay in giving you answers about the interaction energy between molecules and surfaces where the interactions are strong,” said Alfe. If the interaction is weak one must look elsewhere for answers, hence the need for QMC. “QMC is more suited to Jaguar’s architecture,” said Alfe. Furthermore, he added, it is more befitting to leadership-class machines due to its complexity, and while it is currently more expensive, in the future it will be the more common method in conjunction with the advanced speed of modern supercomputers.

Fortunately, QMC and Jaguar make a good match. CASINO is very parallel in nature, regularly running on 10,000 to 20,000 cores with almost perfect scaling, making it much more parallel than its DFT counterparts. In fact, the researchers recently scaled CASINO to 40,000 cores on Jaguar and envision additional improvements that will enable the code to scale to more than 100,000 cores, making it one of the premier codes in the world.

QMC was also relatively easy to port to Jaguar, said Alfe, primarily because the application was previously running on Hector, a Cray XT4 located at the University of Edinburgh. Given the similar platforms, the researchers didn’t even have to recompile their code. And from the looks of things, the team is pleased with Jaguar’s performance.

“Jaguar is a fantastic machine,” said Alfe, “allowing us to do things we couldn’t do without it.”

From our origins to our future, Alfe’s work on Jaguar cuts a wide swath of discovery. Understanding nature’s smallest, most complex systems requires the petascale power of mankind’s most powerful computer, making Alfe’s project and the OLCF’s flagship system a perfect match.