Freeze-and-release direct optimization method for variational calculations of excited electronic states (2501.18568v1)

Abstract: Time-independent, orbital-optimized density functional approaches outperform time-dependent density functional theory (TDDFT) in calculations of excited electronic states involving a large rearrangement of the electron density, such as charge transfer excitations. However, optimizing orbitals for excited states remains challenging, as the latter typically correspond to saddle points on the electronic energy surface. A simple and robust strategy for variational orbital optimization of excited states is presented. The approach involves two steps: (1) a constrained energy minimization, where a subset of orbitals changed by the excitation are frozen, followed by (2) a fully unconstrained saddle point optimization. The constrained minimization step makes it possible to identify the electronic degrees of freedom along which the energy needs to be maximized, preventing variational collapse. Both steps of this freeze-and-release strategy are carried out using direct optimization algorithms with a computational scaling comparable to ground state calculations. Numerical tests using a semilocal functional are performed on intramolecular charge transfer states of organic molecules and intermolecular charge transfer states of molecular dimers. It is shown that the freeze-and-release direct optimization (FR-DO) approach can successfully converge challenging charge transfer states, overcoming limitations of conventional algorithms based on the maximum overlap method, which either collapse to lower energy, charge-delocalized solutions or fail to converge. While FR-DO requires more iterations on average, the overall increase in computational cost is small. For the NH3-F2 dimer, it is found that unlike TDDFT, orbital-optimized calculations reproduce the correct long-range dependency of the energy with respect to the donor-acceptor separation without the need to include exact exchange in the long range.