The one-body reduced density matrix (1RDM) plays a fundamental role in describing and predicting crucial quantum features of bosonic systems, such as the famous Bose-Einstein condensation. The recently proposed reduced density matrix functional theory for bosonic ground states [1] establishes the existence of a universal functional that recovers quantum correlations exactly for all quantum bosonic systems. Based on a novel decomposition of the 1RDM, we have developed a method to design reliable approximations for such universal functionals [2]. This crucial simplification of the so-called constrained-search approach of functional theories allows us to make extensive use of standard machine learning methods to perform a quite efficient computation of such universal functionals and their functional derivative. For the Bose-Hubbard model, we trained a fully connected neural network that outputs a high-quality universal functional. We present a comparison between our approach and the quantum Monte Carlo method. [1] C. L. Benavides-Riveros, J. Wolff, M. Marques, and C. Schilling, Phys Rev. Lett. 124, 180603 (2020). [2] J. Schmidt, M. Fadel, and C. L: Benavides-Riveros, Phys. Rev. Research 3, L032063 (2021).

Download the slides for this talk.Download ( PDF, 7958.71 MB)