# Fully quantum embedding with density functional theory for full configuration interaction quantum Monte Carlo

## TL;DR

This paper introduces a fully quantum embedding approach for i-FCIQMC using a Huzinaga projection, enabling accurate calculations of large systems like molecules with bonds to benzene, overcoming previous convergence issues.

## Contribution

The work develops a novel quantum embedding method for i-FCIQMC that improves convergence and accuracy for large, complex systems, extending its applicability.

## Key findings

- Embedding enables i-FCIQMC to converge to CCSD(T) benchmarks.
- Without embedding, i-FCIQMC struggles with large systems.
- Embedding reduces computational difficulty and improves accuracy.

## Abstract

In common with many high-accuracy electronic structure methods, the initiator adaptation of full configuration interaction quantum Monte Carlo (i-FCIQMC) has difficulty treating realistic systems with large numbers of electrons. This barrier has prevented the application of i-FCIQMC to questions of catalysis that, even for the simplest of models, require high-accuracy modeling of several features of the electronic structure, such as strong and dynamic correlation, and localized vs. delocalized bonding. We here present a fully-quantum embedded version of i-FCIQMC , which we apply to calculate the bond dissociation energy of an ionic bond (LiH) and a covalent bond (HF) physisorbed to a benzene molecule. The embedding is performed using a recently-developed Huzinaga projection operator approach, which affords good synergy with i-FCIQMC by minimizing the number of orbitals in the calculation. We find that, without embedding, i-FCIQMC struggles to converge these calculations due to their substantial system sizes and a lack of error cancellation between reactants and products. With embedding, the i-FCIQMC calculation converges straightforwardly to CCSD(T) benchmarks. Our results suggest that embedded i-FCIQMC will be able treat system sizes well beyond our current reach (even though embedding introduces an error). We discuss how embedding might be improved (and thus the introduced error reduced) using i-FCIQMC energies as benchmarks.