Can CASSCF model the excited states of xanthophyll lutein? A density matrix renormalization group journey

We present an implementation of the complete active space self-consistent field (CASSCF) method that combines the Cholesky decomposition (CD) of the two-electron integrals with the density matrix renormalization group (DMRG) method as a full-CI solver. This approach significantly reduces both the memory requirements and the computational cost of the configuration-interaction and orbital-optimization steps, enabling calculations with significantly larger and more flexible active spaces than standard CASSCF. We apply this methodology to lutein, a xanthophyll carotenoid whose photophysics is governed by a delicate balance between bright and dark low-lying excited states and is notoriously challenging to model. By analyzing potential energy curves along the bond length alternation (BLA) coordinate for the four lowest singlet states using a range of active spaces, we show that the chemically complete CAS(20,20) π space, although adequate for the ground state, leads to an unbalanced description of the excited-state manifold, with the bright state artificially shifted to high energies and the expected sequence of state crossings lost. When an appropriate active space is employed, CD-DMRG-CASSCF captures the evolution of the ground and low-lying excited states along the BLA coordinate, correctly describing changes in electronic character, tracked through transition dipole moments, as well as state crossings, demonstrating that a physically meaningful description arises only when the active space includes higher-lying π* character while avoiding low-lying orbitals that overstabilize dark states. Although the computed excitation energies remain qualitative due to the lack of dynamic correlation, the method reliably reproduces the ordering and nature of the lowest excited states. These results demonstrate both the practical potential and the limitations of large-scale CD-DMRG-CASSCF for conjugated chromophores and motivate future extensions including dynamic correlation and environmental effects.