Excess carbon emissions must be addressed to avoid catastrophic climate impacts. At the moment, just reducing emissions is not enough. Direct air capture, a technology that extracts carbon dioxide from the surrounding air, has great potential to help solve the problem.
However, there are major challenges. Direct air capture technology requires unique designs for all types of environments and locations. For example, the direct air capture configuration in Texas necessarily differs from that in Iceland. These systems must be designed with precise humidity, temperature, and air flow parameters for each location.
Georgia Tech and Meta are now collaborating to create a large database that could make designing and implementing direct air capture technology easier and faster. The open source database allowed the team to train his AI model orders of magnitude faster than existing chemical simulations. The project, called OpenDAC, has the potential to accelerate climate change solutions that the planet desperately needs.
The team's research ACS Central ScienceJournal of the American Chemical Society.
“When capturing air directly, there are many ideas about how to best take advantage of the air flow and temperature fluctuations of a particular environment,” said Dr. author. “But the big challenge is finding materials that can efficiently capture carbon under the specific conditions of each environment.”
Their idea was to “create a database and toolset to broadly assist engineers who need to find the right materials that work,” Medford said. “We wanted to use computing to go from not knowing where to start to giving you a solid list of materials to synthesize and try.”
Containing reaction data for 8,400 different materials and utilizing approximately 40 million quantum mechanical calculations, the researchers believe this dataset is the largest and most robust of its kind.
Building partnerships (and databases)
Researchers from Meta's Fundamental AI Research (FAIR) team were looking for ways to harness the power of machine learning to address climate change. They identified direct air capture as a promising technology and needed to find a partner with expertise in materials chemistry related to carbon capture. They headed straight to Georgia Tech.
ChBE Professor David Sholl, Faculty Fellows Cecile L. and David IJ Wang, and Director of the Transformative Decarbonization Initiative at Oak Ridge National Laboratory, are among the world's top experts in metal-organic frameworks (MOFs). I'm one of the. These are the types of materials that can be expected to directly capture air due to their cage-like structure and proven ability to attract and capture carbon dioxide. Scholl enlisted Medford, who specializes in applying machine learning models to atomic and quantum mechanical simulations relevant to chemistry, to the project.
Mr. Scholl, Mr. Medford, and their students provided all input to the database. Since the database predicts her MOF interactions and the energy output of those interactions, considerable information was required.
They needed to know the structure of nearly every known MOF, both the structure of the MOF itself and the structure of the MOF that interacts with carbon dioxide and water molecules.
“To predict what a material will do, you need to know where all the atoms are and what their chemical elements are,” Medford says. “Understanding the input into the database was half the problem, and that's where our Georgia Tech team brought core expertise.”
The team took advantage of a large collection of MOF structures previously developed by Sholl and his collaborators. They also created a large collection of structures containing defects found in practical materials.
The power of machine learning
Anuroop Sriram, director of research and engineering at FAIR and lead author of the paper, generated the database by running quantum chemical calculations on input provided by the Georgia Tech team. These calculations used approximately 400 million CPU hours. This is hundreds of times more calculations than an average academic computing lab can perform in his year.
FAIR also trained a machine learning model on the database. A machine learning model trained on 40 million calculations was able to accurately predict how thousands of MOFs would interact with carbon dioxide.
The research team demonstrated that the AI model is a powerful new tool for materials discovery, offering comparable accuracy to traditional quantum chemistry calculations, but faster. These features will allow other researchers to extend the study to investigate many other MOFs in the future.
“Our goal is to look at the set of all known MOFs and find the one that most strongly attracts carbon dioxide without attracting other air components such as water vapor, and use these high-precision quantum calculations to do so. It was to be done,” Sriram said. “To our knowledge, this has not been possible with any other carbon capture database.”
Using a proprietary database, the Georgia Tech and Meta teams identified approximately 241 MOFs with a high potential for direct air capture.
move forward with impact
“According to the United Nations and most developed countries, we need to achieve net-zero carbon emissions by 2050,” said Matt Whittender, director of Meta's FAIR chemistry team and co-author of the paper. Mr. Le said. “Most of this has to be achieved by stopping carbon emissions completely, but we also have to address historical carbon emissions and sectors of the economy that are very difficult to decarbonize, such as aviation and heavy industry. That's why CO2 removal technologies like direct air capture need to be available within the next 25 years.”
Direct air capture is still a nascent field, but the researchers say it's important that innovative tools are currently being developed, such as the OpenDAC database made available in the team's paper.
“There is no one solution to achieving net-zero emissions,” Sriram said. “Direct air capture has great potential, but it needs to be scaled up significantly to have a real impact. I think the only way to get there is to find better materials. ”
Researchers from both teams hope the scientific community will join in the search for suitable materials. The entire OpenDAC dataset project is open source, from data to models to algorithms.
“We hope this will accelerate the development of negative emissions technologies that might not otherwise be possible, such as direct air capture,” Medford said. “As a species, we have to solve this problem at some point. I hope that this work can contribute to getting us there, and that there is a real aim to make that happen.” I think.”
Note: Georgia Tech ChBE graduate students Sihoon Choi, Logan Brabson, and Xiaohan Yu made significant contributions and are co-authors of this paper.
QuoteIn: A. Sriram et al., Open DAC 2023 dataset and the challenges of sorbent discovery in direct air capture, ACS Central Science (2024).
Toi: https://doi.org/10.1021/accentsci.3c01629
