Sonali Joshi | Searcharxiv

arXiv2025

Quantum Monte Carlo assessment of embedding for for strongly correlated defects: interplay between mean-field starting point and interactions

Point defects are of interest for many applications, from quantum sensing to modifying bulk properties of materials. Because of their localized orbitals, the electronic states are often strongly correlated, which has led to a proliferation of quantum embedding techniques to treat this correlation. In these techniques, most of the one-body states are treated with a weakly correlated theory such as density functional theory, and certain one-body states are singled out as an active space to be treated using an effective interaction. We assess these techniques for iron and chromium defects in aluminum nitride using quantum Monte Carlo (QMC) calculations on identical Hamiltonians. For these systems, we find the dominant errors in the embedding arise from the one-body crystal-field splitting in the d orbitals inherited from density functional theory (DFT), rather than double counting corrections, with the screened interactions also affected by the DFT orbitals. Strikingly, the best double counting recipe is opposite in these two cases. Because excitation energies can agree while the underlying wave functions do not, diagnosing these errors requires detailed information about the many-body wave functions, which QMC provides.

cond-mat.str-el↗

arXiv2025

Renormalized density matrix downfolding: A rigorous framework in learning emergent models from ab initio many-body calculations

We present a generalized framework, renormalized density matrix downfolding (RDMD), to derive systematically improvable, highly accurate, and nonperturbative effective models from ab initio calculations. This framework moves beyond the common role of ab initio calculations as calculating the parameters of a proposed Hamiltonian. Instead, RDMD provides the capability to decide whether a given effective Hilbert space can be identified from the ab initio data and assess the relative quality of ansatz Hamiltonians. Any method of ab initio solution can be used as a data source, and as the ab initio solutions improve, the resultant model also improves. We demonstrate the framework in an application to the downfolding of a hydrogen chain to a spin model, in which we find the interatomic separations for which a nonperturbative mapping can be made even in the strong coupling regime where standard methods fail, and compute a renormalized spin model Hamiltonian that quantitatively reproduces the ab initio dynamics.

cond-mat.str-el↗