In heterogeneous catalysis, calculations require complicated computational processes due to the variability of reaction pathways and possibilities. Now, thanks to a clever combination of programming and machine learning, researchers have achieved a dramatic increase in the speed of simulations, enhancing the energy efficiency of an otherwise resource-hungry process. The results, reported for reactions to convert carbon dioxide into fuels, could rapidly translate into other industrially relevant reactions, such as depolymerisation and biomass valorisation.
‘We’ve unlocked an understanding unattainable by manual simulations, potentially accelerating discovery by orders of magnitude,’ says Núria López from the Institute of Chemical Research of Catalonia, who led the study. This framework facilitates the prediction of selectivity and reactivity in catalysis, particularly the process of producing long-chain hydrocarbons from syngas – commonly called Fischer–Tropsch. ‘Until now, computational calculations were constrained by tedious manual monitoring of intermediates … as well as hundreds of long-running processes as alternative reaction routes appear,’ she adds. Besides speed, the new programme predicts properties, or ‘observables’, such as selectivity, reaction rates and yield, explains López. ‘This is directly comparable with experimental results, which demonstrates the potential of the programme,’ she adds.
In homogeneous and enzymatic catalysis, the active site is usually constrained to a small set of atoms, explains expert in materials science and simulation Anastassia Alexandrova, from the University of California, Los Angeles, US. ‘In heterogeneous catalysis, the surface is large and often complex, presenting a wide variety of possible binding sites for reagents,’ she says. Additionally, the catalyst’s surface is surprisingly dynamic. As the reaction progresses, it suffers a reconstruction process ‘under the influence of the reactants … creating many different microenvironments for every active site.’ Scanning for possible pathways and networks is ‘unsurmountable in terms of required time and computer resources, and also error prone’ if performed manually, says Alexandrova. ‘This paper presents an important step in the right direction, [studying] the reaction … faster, with the help of machine learning.’
What are AI, machine learning and neural networks?
Artificial intelligence (AI) is an umbrella term often incorrectly used to encompass a variety of connected but simpler processes.
AI is the ability of machines and computer programmes to perform tasks that typically only humans could do, such as reasoning, responding to feedback and decision making.
Generative AI is a newer variant of AI that analyses and detects patterns in training datasets to generate original text, images and videos in response to requests from users. ChatGPT, Microsoft Copilot, Google Gemini and more recently X’s Grok are all examples of chatbots that use generative AI.
Neural networks are an interconnected array of artificial neurons, akin to biological brains, that identify, analyse and learn from statistical patterns in data.
Machine learning is a subset of AI that allows machines to learn from datasets and make predictions based on new data, without programmers explicitly asking it to do so. Machine learning models improve their performance as they receive more data.
Deep learning is an enhanced type of machine learning that uses neural networks with many layers to analyse complex data from very large datasets. Applications of deep learning include speech recognition, image generation and translation.
Large language models or LLMs are a type of deep learning trained on large amounts of data to understand and generate language. LLMs learn patterns in text by predicting the next word in the sequence and these models are now able to write prose, analyse text from the internet and hold dialogues with users.
‘This breakthrough framework automatically maps out and analyses massive, complex chemical reaction networks, previously too large or problematic to handle manually,’ explains Ritesh Kumar, who specialises in science and artificial intelligence at the University of Chicago, US. The system ‘quickly and accurately predicts’ the reactivity of surface catalysts, ‘without a scientist guessing every possible step’, he adds. ‘It really replaces guesswork with intelligent automation, instantly estimating the energy and speed of thousands of steps.’
Whereas traditional density functional theory (DFT) programmes predict up to 500 steps in 100 processing hours, this new solution speeds the search by orders of magnitude – simulating 370,000 possible pathways in a similar time. ‘The speed is impressive – it identifies important reactions at a fraction of the time and cost,’ says Kumar, who will soon start a position at TCG Crest, India. And, maybe most importantly, ‘without the enormous energy consumption typically required by supercomputers’. Besides the benefits to sustainability, the automated algorithms could allow scientists to prioritise precise – and slower – computing tools for critical calculations only, explains Kumar. ‘In processes like Fischer–Tropsch, the number of potential pathways explodes into the hundreds of thousands … with traditional techniques it takes centuries to calculate,’ he says. Now, neural networks could automatically discover reaction pathways and speed up the study of complex catalytic processes.
López explains the similarities in simulating processes across catalysts could catapult the applications of the algorithm in industry applications. After Fischer–Tropsch, the group could calculate other complex reactions, including the valorisation of biomass and plastic recycling. ‘Our objective is transitioning into the rigorous environments of industrial research and development, studying aspects often overlooked in academia such as code security, robustness, sustainability and accessibility,’ says first author Santiago Morandi. ‘This preliminary platform provides a potential bridge between theory and experiments, which will allow for rapid data-driven optimisation of chemical reactors.’
Additional information
No comments yet