Many-particle quantum systems can be completely described by N-body wave functions or density matrices. However, such objects are high-dimensional and extremely difficult both to compute and to apprehend with physical intuition, especially for extended systems. In most applications, though, only a tiny part of the information available in many-particle quantum states is really useful. For instance, the requested output of an electronic structure calculation are often simply the effective forces experienced by the atoms of the molecular system. Such quantities could be obtained at a much lower cost using reliable interatomic potentials. Likewise, collective quasiparticle states (molecular or crystalline orbitals, plasmons, phonons, polarons, excitons, etc.) allow one to describe the properties of many-particle quantum systems with lower-dimensional objects, which are easier to visualize and can be computed accurately enough for most physical systems, by means of effective one-particle (Kohn-Sham, TDDFT, GW, …) or two-particle (Bethe-Salpeter equation, …) models. For these reasons, such states play an essential role in condensed matter physics and materials science.
This workshop will bring together a mix of mathematicians, physicists, chemists, computer scientists, and biologists to address some of the following questions: Can machine learning (ML) techniques be used to create ab-initio accurate interatomic potentials? Can they generate quasiparticle states or approximations thereof given only the molecular Hamiltonian as an input and macroscopic observables as an output? On a larger scale and going towards materials design (materials genomics): how can one generate the necessary and sufficient data to use ML approaches to infer the important collective variables (“materials genes”, scaling relations, etc.)?
This workshop will include a poster session; a request for poster titles will be sent to registered participants in advance of the workshop.