WEMO24-Compiled Report-Interactive compressed (002) 4154-7882-8116 v.1 - Flipbook - Page 61
New catalysts discovery for e-fuel
About e-fuels
E-fuels are synthetic fuels made of carbon dioxide (or monoxide) and
hydrogen, produced with decarbonized energy. They are alternative
solutions, fungible in the energy system, to decarbonize sectors
without alternatives, like long-range aviation. They are part of the
toolbox to achieve short-, medium-, and long-term climate objectives.
PREVIOUS
ARTICLE
NEXT
ARTICLE
Each reasoning step is performed until the LLM reaches an optimal point. It must be noted that experiments will still be required from time to
time to constrain the prediction model and ground the LLM with real-world data. With such an approach, new catalysts may be suggested to
the chemists in a few minutes, a limiting step in real-world experiments.
HOME
The Realistic case
for Gen AI
How Gen AI can be leveraged to accelerate e-fuels development?
Designing better catalysts is critical to lowering the global energy bill required to produce e-fuel and thus its production costs.
As new electrical energy vectors, e-fuels enable indirect electrification,
particularly as it relates to the hardest-to-decarbonize sectors, such
as heavy mobility (aviation and maritime fuels) and the chemical
industry (transformation reagents). Additionally, they facilitate the
recycling of unavoidable CO2 emissions, like those from the cement
industry.
However, those fuels are very expensive to produce, costing up to
100 times more than conventional petroleum1. To decrease those
costs, a number of levers can be activated, especially:
• the decarbonized electricity cost used to capture CO2 and produce
hydrogen
• the catalyst efficiency for the CO2 to syngas reaction or for the
Fischer-Tropsch reaction
1 https://www.pik-potsdam.de/members/Ueckerdt/E-Fuels_Stand-und-Projektionen_PIKPotsdam.pdf
WEMO 2024
One can leverage multi-agent LLM to orchestrate several highly
specialized agents. one generates new formula, a second one predicts
the desired properties for the application, in the case of catalysts,
stability and activation energy) of the generated molecule. Finally,
one can leverage the planning and reasoning ability of the Large
Language Model to analyze the results and propose a new molecule.
Source: https://chemrxiv.org/engage/api-gateway/chemrxiv/assets/orp/resource/item/66220456418a5379b0297f8d/original/large-language-models-open-new-way-of-aiassisted-molecule-design-for-chemists.pdf
61
02
