Micado : Microdroplet catalysis of CO2-amine C-N bond formation, from machine learning-based simulations to predictive control
Description of the PhD project
Carbon–nitrogen (C–N) bond formation from CO₂ and ammonia or amines is central to carbon capture technologies, atmospheric chemistry, and potentially prebiotic chemistry. A landmark recent study by our experimental partner at ETH Zürich (Signorell group) demonstrated that aqueous microdroplets promote the spontaneous formation of urea from CO₂ and ammonia under ambient conditions, a reaction that does not proceed in bulk solution. This striking microdroplet catalysis reveals reactivity fundamentally different from that of homogeneous solution, yet its molecular origin remains unknown. No predictive framework currently exists to describe or control such chemistry in confined aqueous environments.
The MICADO project aims to uncover the molecular mechanisms underlying microdroplet catalysis of CO₂–amine C–N bond formation and to build the first predictive, multiscale model linking droplet physicochemical properties to reaction rates and product distributions. To achieve this, the ENS team will develop machine-learning interatomic potentials (MLIPs) trained at hybrid-DFT accuracy using active learning protocols, building on the group’s ArcaNN automated training framework and state-of the-art foundation models such as MACE-OMOL. These MLIPs will enable large-scale reactive molecular dynamics simulations with quantum accuracy and extensive statistical sampling, which are inaccessible to conventional ab initio approaches.
Reaction mechanisms will be explored using MLIP-based molecular dynamics combined with transition path sampling, enabling unbiased identification of competing pathways (sequential vs. concerted amidation, partial vs. full amidation to urea), free-energy profiles, and the key catalytic factors, including interfacial electric fields, local acidity, confinement, and reactant enrichment. Simulations will be performed in bulk solution, at the air–water interface, and in droplets of varying sizes. The molecular-level information will be integrated into mesoscale reaction–diffusion models to connect atomistic mechanisms to droplet-scale observables such as reaction kinetics and product distributions.
The project follows a strict theory–experiment synergy : controlled single-droplet measurements from the ignorell group at ETH Zürich (covering reaction kinetics, product identification, and systematic variation of droplet conditions) will serve as benchmarks to validate and refine the models. Discrepancies will guide targeted improvements of the ML potentials and identification of missing physical ingredients. This iterative loop between simulation and experiment is essential to establish a truly predictive framework.
Beyond the CO₂–NH₃ system, the project will extend to CO₂–amine reactions in the presence of atmospherically relevant organic acids, providing design principles for interfacial catalysis more broadly. The methodology, combining foundation-model fine-tuning, active learning, and automated mechanism discovery, will be transferable to other chemically reactive systems. The project will demonstrate how AI-based molecular simulation can transform the understanding and prediction of complex interfacial reactivity, with implications for carbon capture, atmospheric chemistry, and prebiotic processes.
Keywords
Theoretical chemistry | Machine learning | Raman spectroscopy | Microdroplets | CO2 transformation
Research Unit, UMR number and acronym
ENS - Laboratoire de Chimie Physique et Chimie du Vivant | UMR 8228 CPCV
Description of the research Unit/subunit
The Theoretical Chemistry group at the CPCV laboratory (ENS-PSL, UMR 8228) holds a leading international position at the interface of artificial intelligence and chemistry. Over the past three years, the group has produced more than twenty publications in this area, including papers in Nature Chemistry, Science, JACS, PNAS, Nature Communications, and Digital Discovery. Research axes include machine-learning interatomic potentials for reactive systems, neural-network molecular dynamics, reaction mechanism discovery, and IA-driven enhanced sampling. The group developed the ArcaNN automated workflow for generating training sets for reactive MLIPs, which will be central to this PhD project. The group benefits from an exceptional local ecosystem within ENS-PSL, which hosts the Centre for Data Science and coordinates the ChemAI PSL program of which the CPCV group is a founding member.
Name of the supervisor
Damien Laage (damien.laage@ens.psl.eu)
Name of the co-supervisor
Ruth Signorell (ruth.signorell@phys.chem.ethz.ch)
3i Aspects of the proposal
Intersectoriality
The MICADO project addresses two challenges of major industrial and societal relevance : CO₂ capture and CO₂ valorization. The conversion of CO₂ into value-added nitrogen-containing compounds, such as urea and its derivatives, is of direct interest to the chemical industry, notably for fertilizer synthesis and polymer production. Current industrial urea synthesis (the Bosch–Meiser process) requires high pressures and temperatures ; microdroplet-based approaches operating under ambient conditions could open radically more energy-efficient routes. Beyond urea, a general predictive understanding of CO₂–amine reactivity in confined aqueous environments will guide the design of novel CO₂ scrubbing solvents and droplet-based reactors. The predictive multiscale model developed in this project will provide the type of quantitative design framework required for innovation in industrial carbon capture and utilization. The project thus sits at the intersection of fundamental research and potential technological applications of high value for the energy transition.
International
The project is built around a close collaboration with the group of Prof. Ruth Signorell at ETH Zürich, one of the world’s leading laboratories in the physical chemistry of individual aerosol particles and droplets. The Signorell group recently reported the spontaneous formation of urea from CO₂ and ammonia in aqueous microdroplets (Science, 2025), a result that directly motivates and anchors this PhD project. The PhD student will carry out a research stay of at least one month at ETH Zürich, where they will work alongside the experimental team. This stay is scientifically integral to the project : exposure to single-droplet trapping techniques, Raman spectroscopy, and controlled variation of droplet conditions will directly inform the choice of simulation parameters and provide the experimental benchmarks required to validate the multiscale models developed at ENS. Close interaction with the ETH group will be maintained throughout the PhD via regular joint meetings and exchanges.
Interdisciplinarity
MICADO tightly integrates three disciplines : theoretical chemistry, artificial intelligence, and experimental physical chemistry. On the theoretical side, the project requires expertise in quantum chemistry (for training data generation), statistical mechanics (for enhanced sampling and free-energy methods), and chemical kinetics (for multiscale reaction–diffusion modeling). The AI dimension involves the development and fine-tuning of machine-learning interatomic potentials, novel active learning strategies, foundation model adaptation for reactive systems, and generative AI, which are methods at the frontier of current AI research applied to the molecular and materials sciences. The experimental dimension, provided by the Signorell group at ETH Zürich, brings single-particle trapping techniques, aerosol spectroscopy, and quantitative reaction kinetics in isolated droplets. This three way integration is an absolute requirement to address this complex problem. Neither simulation alone nor experiment alone can resolve the molecular mechanisms of microdroplet catalysis ; only their tight coupling through an AI-driven multiscale framework can do so.
Expected profile of the candidate
We are looking for a candidate holding a Master’s degree in theoretical chemistry, computational chemistry, or a closely related field such as physical chemistry or chemical physics. A solid background in quantum chemistry and/or statistical mechanics is expected, as well as familiarity with molecular simulation methods (molecular dynamics, Monte Carlo, or related techniques). Experience with machine-learning interatomic potentials (MLIPs) or other AI/ML approaches applied to molecular systems is considered an asset but is not required ; the PhD project will provide comprehensive training in these areas. Programming skills are expected. The candidate should demonstrate scientific curiosity, rigor, and motivation to work at the interface of AI and chemistry. We are committed to fostering an inclusive and diverse academic community and encourage applications from all backgrounds.
Important dates
Call for applications : from September 1st to October 31st 2026
Eligibility check results : November
3i Committee evaluation results : December
Interviews from the shortlisted candidates with the Selection Committee : January 2027
Start of the PhD : March 1st 2027