New Multireference Quantum Chemistry Methods Based on Renormalized States

The efficient compression of Hilbert space is one of the most crucial problems in modern quantum simulation of strongly correlated systems. One approach involves using isolated states obtained within the system, such as energetically low-lying adiabatic states, to describe the Hilbert space of the system, but without considering the interaction between the system and its environment. According to quantum information theory, Schmidt decomposition can produce two sets of the most compact states representing the bipartitioned Hilbert space (system and environment) of a particular wave function. However, such procedures require knowledge of the total wave function and are hindered by the enormous degrees of freedom in the environment.
My collaborators and I propose performing the decomposition within a model system composed of the system of interest and a defined bath space, compressed from the real environment space, to obtain renormalized states. In this talk, I will introduce two approaches based on this idea--renormalized-residue-based multireference configuration interaction (RR-MRCI) and block interaction product state density matrix renormalization group (BIPS-DMRG). RR-MRCI has demonstrated superior performance in recovering dynamic correlation, achieving energy errors within 1 milliHartree (mEh} relative to uncontracted MRCI, and has been successfully applied to molecules with active spaces of around 30 orbitals. BIPS-DMRG is designed to handle static correlation in larger systems with inhomogeneous correlations, such as polymers and molecular aggregates, based on local correlation approximation. Benchmarking tests on systems with active spaces ranging from (36e, 360) to (100e, 1000) have shown that BIPS-DMRG produces accurate results, with errors of 1-3 mEh. I will show the improved results of these two methods compared with standard methods using isolated states, including internally contracted MRCI (ic-MRCI) and active space decomposition (ASD)-DMRG. Additionally, I will demonstrate the spin adaptation can be implemented in these two methods by considering the coupling coefficients during the tensor contraction.