How Machine Learning Potentials Unlock Realistic Modeling and Design of Nanoporous Supercapacitors
by Zacharie Waysenson
PhD, Laboratoire de Physique de l’École Normale Supérieure (Paris)
The next AI4Science seminar is coming soon !
AI Factory France is pleased to announce the 5th “AI for Science” seminar, organized by SCAI.
The topic of the fifth session is “How Machine Learning Potentials Unlock Realistic Modeling and Design of Nanoporous Supercapacitors” – by Zacharie Waysenson, PhD at Laboratoire de Physique de l’École Normale Supérieure of Paris.
To participate at the seminar, the registration is mandatory.
Abstract:
Electrochemical Double-Layer Capacitors (EDLCs) play a pivotal role in the transition to renewable energy by bridging the gap between high-energy batteries and high-power dielectric capacitors. However, the computational discovery and design of carbon electrodes have long been bottlenecked by model approximations and idealized pore geometries. Real-world porous carbons are mechanically compliant, topologically disordered networks whose performance is governed
by an intricate interplay between sub-nanometer confinement, desolvation barriers, and dynamic electromechanical feedback. Machine learning interatomic potentials (MLIPs) provide the tools needed to bridge this accuracy–scale divide, enabling quantum-accurate modeling across the length and time scales of electrochemical interfaces. In this work, we show how coupling Atomic Cluster Expansion (ACE) potentials with variational constant-potential molecular dynamics (ConP-MD) resolves predictive gaps and provides actionable physical principles for supercapacitor materials design.
First, we study the role of host framework compliance in out-of-equilibrium charging dynamics. By capturing explicit lattice vibrations and collective breathing modes of the carbon host, the flexible MLIP framework reveals how dynamic wall compliance lightens local steric overcrowding. This mechanism accelerates ion diffusion pathways, driving a kinetic acceleration in charging times compared to conventional rigid-lattice models without compromising integral capacitance [1].
Second, we extend this methodology to large-scale, topologically realistic disordered carbons synthesized via computational soft-templating quenches [3]. By explicitly capturing structural curvature, ring-size distributions, and local
heterogeneity, our simulations reproduce the experimentally observed anomalous capacitance surge in sub-nanometer pores, rationalizing it through confinement effects, cross-pore electrostatic correlations and asymmetric ion desolvation profiles [2].
Finally, we show how merging MLIP-enabled atomistic simulations with emerging topological descriptors and quantum-electronic response models opens the path toward the design of targeted carbon architectures for energy technologies.
[1] Waysenson, Z., France-Lanord, A., Serva, A., Simon, P., Salanne, M., & Saitta, A. M. (2025). Electrode flexibility enhances electrolyte dynamics during supercapacitor charging. ACS Nano.
[2] Waysenson, Z., France-Lanord, A., Simon, P., Salanne, M., & Saitta, A. M. (2026)
Soft-templated carbon models elucidate the anomalous increase of capacitance in nanopores. ACS Nano
[2] Waysenson, Z., Barthe-Golde B., France-Lanord, A., Salanne, M., & Saitta, A. M. (2026)
Quenches methods for templated porous carbon , J. Phys Chem, in publication (2026)
Short bio:
Zacharie Waysenson recently earned his PhD at the Laboratoire de Physique de l’École Normale Supérieure in Paris. His research investigates electrochemical systems at the molecular level, with a specialized focus on the behavior of nanoporous electrodes.
———————————————
“How Machine Learning Potentials Unlock Realistic Modeling and Design of Nanoporous Supercapacitors”
by Zacharie Waysenson, PhD, Laboratoire de Physique de l’École Normale Supérieure (Paris)
- Tuesday, November 10, 2026, 11:00 am – 12:00 pm (Paris time)
- Amphithéâtre Durand, Esclangon Building – Sorbonne Université, 75005 Paris
- Hybrid format – https://visio.numerique.gouv.fr/dst-utys-juy
Supported by



















