Pittsburgh Supercomputing Center 

Advancing the state-of-the-art in high-performance computing,
communications and data analytics.

Building a Better Carbon Trap

Blacklight Helps Researchers Develop Better Materials for Carbon Capture

Read the University of South Florida press release.

In the aftermath of the infamous "Climategate" leak of scientists' emails, it’s becoming clear that the climate science never really was in question. Human generation of carbon dioxide and other greenhouse gasses is altering the global climate. Which leaves us with a difficult question. Today’s world is very dependent on carbon-dioxide-generating fossil fuels. How do we make our economy “carbon neutral” while still having an economy?

Technical Note

Harnessing Memory to Make the Trap Work

Brian Space and his collaborators needed a number of supercomputing resources to move from modeling the behavior of pairs and triplets of gas molecules to bulk properties of gaseous mixtures and their interactions with candidate metal organic framework (MOF) matrices. Perhaps not surprisingly, each of these modeling steps required a different type of computational resource.

“For the different pieces of the puzzle we needed different machines,” Space explains. “After our group develops an in-depth model of the forces between various guest molecules and their host materials, we perform molecular simulations of the sorption-mediated processes on highly parallel [supercomputing ] resources.” His group used a number of machines in the National Science Foundation’s XSEDE network, including TACC Ranger, SDSC Trestles, and GPU clusters such as Georgia Tech’s Keeneland for the latter calculations.

But the initial simulation of small numbers of molecules, requiring exquisitely detailed quantum mechanical modeling, is so memory intensive that only one XSEDE resource — and one program — made sense: PSC Blacklight, running Molpro.

“Their simulations require a lot of memory, so they’re ideal for Blacklight’s shared-memory architecture,” says Marcela Madrid, senior scientific specialist at PSC. “Some of their runs took about a terabyte of memory; one of the problems was to tune the job script so that you would really get the memory you want, and not get errors due to insufficient memory.”

The ability to run Molpro on Blacklight was also important, Space adds. “Other programs would have taken a year or two to perform the calculations; with Molpro, it only took hours.”

“One issue was that Molpro is very I/O intensive,” Madrid says. “We had to efficiently use the file system in order to accommodate this.”

Madrid and Rick Costa, who is a staff computational science consultant at PSC, were instrumental in making it all work, Space says. “Without their help, we couldn’t have done it.”

“Science is really a cooperative effort,” he adds. “Many years of people’s work went into writing the codes to do these calculations at all, let alone efficiently. No one group can be an island; you have to depend on others.”

No one can say yet what technologies we’ll need to solve the climate change dilemma. But capturing carbon dioxide from smokestacks and other waste streams is attractive. This is because it would allow us to continue using fossil fuels, at least for a while.

In a March 7 paper in Nature, a team led by Patrick Nugent of the University of South Florida’s Department of Chemistry and Youssef Belmabkhout of the King Abdullah University of Science and Technology’s Advanced Membranes & Porous Materials Center has reported a new material that may fit the bill for carbon capture. They have developed a series of metal organic frameworks (MOFs) that may be able to “filter” carbon dioxide out of smokestacks and exhaust pipes.

“The new materials show very selective carbon dioxide capture,” says co-author Brian Space, professor of chemistry at the University of South Florida. “Indeed, carbon dioxide fits [into the MOFs] like a glove; nothing else fits as well.”

Space’s team used supercomputers to simulate how passing through an MOF affects exhaust gases. These simulations allowed them to explain the behavior of MOFs and give their lab-researcher colleagues the clues they needed to create better MOF “carbon scrubbers.”

Pittsburgh Supercomputing Center’s Blacklight machine, Space says, was crucial for an initial, detailed series of simulations. These showed how previous simulations had missed something important. The new simulations also pointed to the way forward.

Carbon dioxide molecules (gray and red) trapped in the MOF matrix.

Details count

Space’s group attacked the simulation in steps. The many gases in an exhaust flow and the components of the MOFs could be simulated accurately only after the investigators understood the details of their interactions. And previous models were missing something. They didn’t make good predictions of which MOFs would attract which gases.

Space and his team produced exact simulations of individual gas molecules, and tested how they interacted in pairs and triplets.

Moving from simulating pairs of molecules — which earlier investigators had done — to triplets really increased the demand on computer resources, Space says. “It’s difficult to calculate.” But the researchers discovered that adding the third molecule made a big difference in the computers’ predictions. And it made them right.

Inside the MOFs, Space and his team discovered, electrical charges were distorting the gas molecules, attracting them much like a magnet attracts iron filings. Previous simulations hadn’t included such “polarization,” partly because two-molecule models had suggested it wouldn’t be important. But when the simulation team added the third molecule, they found that polarization’s effect was much stronger than expected.

Only Blacklight, which is the largest “shared memory” computer in the world, had the memory capacity to perform the triplet calculations. (See Technical Note, right)

 “These energy models were not even possible a few years ago,” Space says. “We can now explain the experiments at an unprecedented level of accuracy.”

Give-and-take toward a better carbon trap

Space’s group worked with the laboratory researchers in a give-and-take way.

 “We start out by asking, ‘Why does this work this way?’” he says. “Then we ask, ‘How do you improve this — how would you change the way it works?’”

With the improved simulations, Space’s team could better explain the properties of known MOFs. This knowledge allowed them to experiment — in the simulations — with different MOF structures to see if they would bind carbon dioxide better. The lab researchers created these MOFs, and measured their real properties. Space and his team then plugged differences between the predicted and measured properties into their simulations. In the end, the back-and-forth led to even better predictions.

The work identified a family of MOF structures containing an electrically charged silicon-fluoride compound that attracts carbon dioxide much more strongly than other gases. They even work in the presence of water vapor, which prevented efficient carbon capture in earlier materials. This discovery was important enough to merit an article in Nature, one of the world’s most pre-eminent research journals.

“It’s an exciting time in the field,” he adds. “The level of accuracy, if it’s done carefully, is really predictive now.”