Embedded Neural Quantum State

There is always demand, and especially in light of recent events,  in faster and cheaper methods for simulation of chemical systems. Due to extremely high complexity of the full quantum many-body problem, one needs always to find a compromise between accuracy and computational cost.

At the present time my project in UZH (group of Prof. Dr. Luber) is related to development of quantum chemistry methods for simulation of the "embedded" molecular systems , i.e. interacting with the environment (e.g. molecules in solvent). The environment in this case is represented by less accurate method (e.g. density functional theory - DFT) and the molecule is by accurate many-body method (e.g. DMRG, Quantum Monte Carlo). One of the last developments in these area are the embedded DMRG-in-DFT case (M. Reiher 2015). My main idea is to follow this scheme and to use the new many-body Neural Quantum State (NQS, G. Carleo, 2017) methods instead of DMRG. The development of NQS part is planned to do in collaboration with Dr. Dmitry Bazhanov (Moscow State University - MSU, Moscow, Russia), where such method is currently developed.

In short about NQS. In last years the significant progress was achieved in development of methods in Machine Learning, Neural Networks and its application for simulation of quantum many body systems. The NQS method is based on stochastic simulation principle of a quantum system by using neural networks for representation of a quantum state.  Comparing to the existing methods for simulation of many-body systems like DMRG or "classic" Quantum Monte Carlo family, the new method can potentially take into account higher degrees of correlation in the system requiring less computational resources (first of all memory). The last work devoted to NQS (G. Carleo, 2020) showed that this method can be successfully used for electron system, which makes it possible to apply the NQS for studies of quantum chemical systems. Therefore, it connects this method directly with cheaper practical application in the area of chemistry and, in particular, in drug development.

The technical side of the project is based on code development in frames of the quantum chemistry simulation tool CP2K (DFT, Hartree-Fock). The many-body accurate embedded simulations will be performed using NQS code developed in collaboration with MSU, Russia  and compared with existing NQS code NetKet (G. Carleo) and DMRG code Block. During the collaboration work several visits are planned: Two visits of collaborator from MSU, Moscow in UZH, Zurich and two my visits from UZH Zurich to MSU, Moscow.

As a result of the collaboration the new method "embedded DFT-in-NQS" will be developed and tested. The results will be published in Open-Access journals.


University of Zurich

Dr. Ilia Sivkov

Moscow State University

Dr. Dmitry Bazhanov