Kraken will be decommissioned on April 30, 2014. For more information see Kraken Decommission FAQs
Kraken will be decommissioned on April 30, 2014. For more information see Kraken Decommission FAQs
The National Institute for Computational Sciences

A Compound Problem

Kraken super expedites models, materials, technology

by Gregory Scott Jones

 

 


Few things drive innovation more so than materials. From superconductors to more fuel efficient cars to better batteries, materials are very often the bottleneck that makes or breaks progress.

In order for countless fields of technology to progress, new materials have to be developed and manufactured economically. It’s a difficult endeavor, one that encompasses the entire scientific cycle from theory to experiment. Luckily, materials researchers have a valuable ally in today’s most powerful computers.

Take Kraken for example. Capable of a peak performance of more than one petaflop (or a thousand trillion calculations per second) and managed by the University of Tennessee’s National Institute for Computational Sciences (NICS), Kraken is one of the integrated digital resources of the eXtreme Science and Engineering Discovery Environment (XSEDE), successor to the National Science Foundation’s TeraGrid project.

The Cray XT5, named after a mythical sea monster, is used by researchers whose interests span the scientific spectrum, from astrophysics to climate and weather prediction to, well, the modeling of chemical compounds for the development of novel materials, the same materials that promise to revolutionize technology.

Dr. Stefano Curtarolo of Duke University is a great example. He is part of a team harnessing the power of Kraken to model more than 400,000 chemical compounds, or combinations of elements that via chemical reactions can be separated back into their respective elements, in hopes of discovering new material systems or unknown properties that exist in currently used materials that have gone unnoticed.

“People know certain materials are great at certain things, but they might also be good at something else,” said Curtarolo.

The data for the compounds is derived from experiment and observation. However, only simulation can process and classify such a large number of compounds for a reasonable amount of money in a reasonable amount of time. Even if the compounds themselves are relatively cheap, said Curtarolo, experiments will likely still be expensive. And if the compounds are expensive, as they are likely to be, experiments will be even more expensive yet. With Kraken, the compounds are free, and better yet, the outcomes are fast.

However, all of the compounds are unique. “Some of the natural compounds are very, very big and are made up of many, many atoms,” said Curtarolo. “The more complex the compound, the more difficult the calculation.”

Getting Descriptive Ultimately, Curtarolo’s team models a compound on Kraken to attain its wave function, or a mathematical probability of how a compound’s particles are behaving at different moments in time. “The wave function contains everything,” he said.

From there the wave function data is transferred to the team’s database at aflowlib.org, a consortium managed by Curtarolo and five other lead researchers, which contains descriptors—empirical quantities connecting the calculated microscopic parameters with the macroscopic properties—that look for novel properties in the modeled compounds.

“A database without descriptors is just a sterile set of calculations without any soul. The descriptor is what tells you good what’s good or bad. That’s what tells you about a particular material or phenomenon,” said Curtarolo.

Through the database, researchers can discover two things: novel materials or previously unknown phenomena exhibited by known materials. For example, it was well known for years that a more complex crystalline structure was necessary for superior thermoelectrics, or materials that convert changes in temperature to electricity. Thanks to Curtarolo’s group, researchers now know why and can quantify the effectiveness of a thermoelectric material based on its crystalline structure.

The team has also proposed some phenomena that should exist and actually found them. For instance, it was Curtarolo’s group that proposed the idea of lithium boron as a superconductor. The team also solved the problem of nonproportiality for scintillators, or materials that can detect radioactivity. Some of these materials are better than others and Curtarolo’s team discovered the mechanism, research that was verified experimentally.

The list goes on and on, such as a group of cesium compounds that turned out to be topological insulators, or materials that behave as insulators on the interior while still allowing the movement of metallic charges at the surface. But Curtarolo admits that up until now the team has been better at describing phenomena than finding novel materials. “But it’s just a matter of time” until they catch up.”

All of these discoveries are the result of mega-database scanning. The team is currently creating the infrastructure for the database to be distributed, where anyone can download it and explore the entries. “In approximately one year it will be completely distributed . . . we share what we have,” he said.

The project is currently on its third year on Kraken. The runs are relatively small, from 64 to 128 cores, but that’s an asset when it comes to the database. “It’s better to have small jobs running longer than big jobs running shorter to maximize our throughput,” said Curtarolo.

The potential for the team’s research is difficult to quantify, but one thing is for sure: as more compounds are modeled the potential for transformational materials discoveries will be greatly accelerated, bringing tomorrow’s technology within reach.

About NICS: The National Institute for Computational Sciences (NICS) is a joint effort of the University of Tennessee and Oak Ridge National Laboratory that is funded by the National Science Foundation (NSF). Located on the campus of Oak Ridge National Laboratory, NICS is a major partner in NSF’s Extreme Science and Engineering Discovery Environment (XSEDE).