Machine learning predicts thermal capacities of

image: Organometallic frameworks capturing CO2 from combustion gases.
see After

Credit: SM Moosavi/EPFL

Metalorganic frameworks (MOFs) are a class of materials that contain nano-sized pores. These pores give MOFs record internal surface areas, making them extremely versatile for a number of applications: separating petrochemicals and gases, mimicking DNA, producing hydrogen, and removing heavy metals, anions fluorides and even water gold are just a few. examples.

MOFs are the focus of Professor Berend Smit’s research at EPFL’s School of Basic Sciences, where his group is using machine learning to make breakthroughs in the discovery, design, and even categorization of ever-increasing MOFs. that currently flood chemical databases.

In a new study, Smit and his colleagues developed a machine learning model that predicts the heat capacity of MOFs. “It’s very classical thermodynamics,” says Smit. “How much energy does it take to heat a material one degree? Until now, all engineering calculations have assumed that all MOFs have the same heat capacity, for the simple reason that there is hardly any data available. Seyed Mohamad Moosavi, postdoc within the Smit group, adds: “If there is no data, how can we make a machine learning model? It seems impossible!

The answer is the most innovative aspect of the work: a machine learning model that predicts how the local chemical environment changes the vibrations of each atom in an MOF molecule. “These vibrations may be related to heat capacity,” says Smit. “Before, a very expensive quantum calculation gave us a single heat capacity for a single material, but now we get up to 200 data points on these vibrations. So, by performing 200 expensive calculations, we had 40,000 data points to train the model of how these vibrations depend on their chemical environment.

The researchers then tested their model against experimental data as a verification under real-world conditions. “The results were surprisingly poor,” says Smit, “until we realized that these experiments had been done with MOFs that had solvent in their pores. So we resynthesized some MOFs and carefully removed the synthesis solvent – measured their heat capacity – and the results were in very good agreement with the predictions of our model!

“Our research shows how artificial intelligence (AI) can accelerate problem solving at multiple scales,” Moosavi says. AI allows us to think about our problems in new ways and sometimes even solve them. »

To demonstrate the real impact of the work, engineers at Heriot-Watt University simulated the performance of MOFs in a carbon capture plant. “We used quantum molecular simulations, machine learning, and chemical engineering in the process simulations,” says Smit. “Results showed that with correct MOF heat capacity values, the overall energy cost of the carbon capture process can be much lower than we originally assumed. Our work is a true multi-scale effort, with an enormous impact on the technical and economic viability of the solutions currently envisaged to combat climate change.

Other contributors

  • Free University of Berlin
  • University of Cambridge
  • Heriot Watt University
  • The University of Manchester

Reference

Seyed Mohamad Moosavi, Balázs Álmos Novotny, Daniele Ongari, Elias Mubarak, Mehrdad Asgari, Özge Kadioglu, Charithea Charalambous, Andres Ortega-Guerrero, Amir H. Farmahini, Lev Sarkisov, Susana Garcia, Frank Noé, Berend Smit. A data science-based approach to predicting the heat capacity of nanoporous materials. Nature Materials October 13, 2022. DOI: 10.1038/s41563-022-01374-3


Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of press releases posted on EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.

Sherry J. Basler