Joining forces for biofuels
Computing and neutron science overcome bioenergy roadblocks
Collaborating with scientists from ORNL's Bioenergy Science Center, Smith focuses on a particularly troublesome molecule called lignin. Lignin is a major component of plant cell walls, where it intertwines with plant sugars called cellulose and hemicellulose. Lignin's hardiness provides benefits to plants in the wild but has been a major frustration to bioenergy researchers. During biofuel production, in which plant biomass is converted into alcohol, lignin inhibits the release of sugars necessary for fermentation.
This recalcitrance is a major bottleneck in the production of biofuels from non-food crops such as switchgrass and poplar trees. Smith's team of researchers has been trying to understand how and why lignin behaves as it does, which has proven to be no easy task.
"One of the problems with lignin is that it doesn't seem to have a sequence," Smith says. "If you think of DNA, it has a code, a sequence of letters that makes up your genome, what you are. Lignin is a mixture of chemicals, but in a seemingly random order; there's no sequence to them.
"There is less known about lignin than almost any other biopolymer because it's only found in plants, which makes it less interesting to medicine. It also has a complex random structure, making it difficult to apply many techniques that you normally would to understand polymers and biopolymers."
To figure out the physical basis of lignin's recalcitrance, Smith combined neutron-scattering experiments at ORNL's neutron facilities with high-performance simulations on lab supercomputers. Using the two resources in tandem required a large amount of theoretical research and technique development to allow experiments to be translated into simulations.
"Neutron experiments tend to produce relatively simple and smooth-looking signals, as they only 'see' a molecule's motions at low resolution," Smith says. "In contrast, data from a supercomputer simulation are complex and difficult to analyze, as the atoms move around in the simulation in a multitude of jumps, wiggles and jiggles. Reconciling these different views of the same phenomenon is a tricky problem."
The complementary techniques of simulation on ORNL's Jaguar supercomputer and neutron scattering at the lab's High Flux Isotope Reactor enabled Smith's team, working with experimentalists from the ORNL Biofuels Science Focus Area, to resolve the structure of lignin aggregates down to 1 angstrom—that's one 10-billionth of a meter, smaller than the width of a carbon atom. These aggregates, known as "clumps," cause problems in biofuel production because they latch on to enzymes used in pretreatment to release sugars from cellulosic biomass.
The combination of experiments and simulation yielded a molecular model of lignin's surface structure at scales ranging from 1 to 1,000 angstroms, giving the research team a broad yet highly detailed view of the molecule. Additional supercomputing simulations revealed how lignin's structure changes during the high-temperature pretreatment process. Smith's team determined that lignin collapses to form problematic clumps even at relatively hot temperatures, rather than only during the cool-down phase, as previously believed.
"Looking at the simulations and experiments together, we've been able to figure out when lignin scrunches up into a ball and when it likes to float around in an extended form," Smith says. "That's very important to know, and it looks like we have a consistent picture. When lignin collapses and when it extends, under what circumstances, and how it interacts, what drives it to interact with cellulose or otherwise—this basic understanding is needed for improving the biofuel production process. It requires both the neutrons and the simulations."
As ORNL's flagship supercomputer, Jaguar, morphs into the more powerful Titan, Smith is looking forward to more advanced simulations of bioenergy-related systems beyond lignin.
"Titan allows us to perform simulations of bigger systems," Smith says. "Now we'll be thinking not just of simulations of lignin and cellulose, but we'll be including enzymes and even the microbes—the things that eat up the biomass. We'll be thinking of the interface between the plant surface and microbe surface and maybe even doing simulations of the whole interface at atomic detail."
Smith also anticipates the need for software and methods that must be developed to keep up with the increase in supercomputing speed.
"As supercomputers increased in power, until about 5 years ago, it was fairly simple to just take your program and run it on the faster computer," he explains. "And it would run faster. But that's all changing now. Because the architecture of these computers is becoming so complex, it leads to huge challenges in getting these programs to run quickly.
"When we get to exascale, in principle, we'll be able to simulate a whole living cell at atomic detail—every atom in a cell will be in our simulation. In practice, that's not going to happen without tremendous methods development."
The full benefits of these pioneering advances in supercomputing may not be realized for years, Smith says, referencing the first molecular simulations, which were made in the late 1970s on computing systems a hundred million times less powerful than those currently available. "Those first simulations were not necessarily very scientifically useful at the time, but they showed you could do it," he says.
"Then later everyone else caught up, and you had this massive amount of information available. In a way, you're putting a marker down; you're saying that we're going here, in terms of system size or level of detail, and proving you can do it. Someone else may come along afterwards and extract even more interesting science, using our methods or following in our footsteps."
Although Smith's research team has made great strides in joining the power of supercomputers with neutron experiments, there is still work to be done. Smith envisions a future in which a scientist who visits ORNL to conduct experiments at the Spallation Neutron Source will have simultaneous access to supercomputing facilities and assistance in using them. Part of this process involves training neutron researchers to design their experiments in such a way that the results can be more easily interpreted using simulations.
"In the future, this approach is going to extend all over the sciences—materials science, bioenergy, condensed matter physics—most of the types of science that SNS deals with," Smith says. "They already use high-performance simulation at SNS on a regular basis to interpret results. But the question for the future is whether Oak Ridge is going to put together a framework for making this nexus between supercomputing and facilities at SNS such that the users have a really unique experience." —Morgan McCorkle