Machine Learning for the Many-Electron Problem
Giuseppe Carleo (EPFL, Lausanne)
| When |
Dec 01, 2025
from 11:00 to 12:00 |
|---|---|
| Where | Salle Condorcet |
| Contact Name | Tommaso Roscilde |
| Attendees |
Giuseppe Carleo |
| Add event to calendar |
vCal
iCal
|
The quantum many-body problem underlies our theoretical description of correlated electrons in molecules, materials, and ultracold atomic systems. Its intrinsic complexity arises from the exponential growth of the Hilbert space with system size, which has long motivated the development of approximate wave-function ansätze and advanced Monte Carlo methods. Neural-network quantum states provide a new class of variational representations capable of capturing complex correlations beyond traditional forms.
In this colloquium, I will begin with a general introduction to the many-body problem and to the use of machine-learning architectures in quantum physics. I will then discuss recent advances in neural-network approaches for ab-initio systems, focusing on message-passing neural-network ansätze for electrons in continuous space. These methods reach high accuracy in benchmark problems such as the homogeneous electron gas and extend the range of accessible system sizes beyond previous neural-network architectures. I will also present a Pfaffian-based neural quantum state for ultracold Fermi gases, which achieves state-of-the-art accuracy across the BCS–BEC crossover. Finally, I will outline ongoing developments aimed at extending neural ansätze to real-time quantum dynamics and finite-temperature properties.
References
—
[1] Carleo and Troyer, Science 355, 602 (2017).
[2] Hermann et al., Nat. Rev. Chem. 7, 692 (2023).
[3] Pescia et al., Phys. Rev. B 110, 035108 (2024).
[4] Kim et al., Commun. Phys. 7, 148 (2024).
[5] Nys, Pescia, Sinibaldi, and Carleo, Nat. Commun. 15, 9404 (2024).