Sijia S. Dong

Multireference calculations are well-suited for describing properties of highly correlated systems. In order to realize the full potential of multireference methods, it is necessary to judiciously choose the active space. It is possible to do this manually, but an automatic procedure for selecting the active space can benefit high-throughput applications and is not susceptible to human error. We previously developed two protocols, GDM-AS and EDM-AS, where the dipole moment calculated at the complete active space self-consistent field (CASSCF) level of theory is used to select the active space, and demonstrated the effectiveness of these protocols when the chosen active spaces are used to find low-lying excitation energies for molecules with nonzero dipole moments. In this work, we demonstrate that using complete active space configuration interaction (CASCI) dipole moments, with the orbitals from the second-order Møller–Plesset perturbation theory, instead of CASSCF dipole moments in our dipole moment active space selection (DM-AS) protocols can significantly improve the efficiency of active space selection compared to our previous methods while maintaining reasonable complete active space second-order perturbation theory (CASPT2) and complete active space pair-density functional theory (CAS-PDFT) vertical excitation energy. We also present new protocols, CASCI-D2DM-AS and its directional version, designed for charge transfer states. We discuss the performance of our new protocols in their ability to describe excitation energy of not only small organic molecules tested in our previous work, but also larger conjugated polar molecules, nonpolar molecules, and bimolecular complexes, to describe the ground-state spin state of transition metal oxides, and to describe bond dissociation curves of small molecules.

The exploration of potential energy surfaces (PESs) is central to the understanding of ground- and excited-state reactions. The construction of potential energy surfaces with multireference methods are usually done sequentially, meaning the calculation usually starts from one geometry, and the molecular orbitals and configuration interaction coefficients are carried over to a successive geometry close to the previous geometry, and so on, ensuring fast and reliable convergence to the desired solution. While this likely yields smooth, reasonable PESs, it is slow because each calculation must wait for the previous one to finish. When multiple processors are available, multiple geometries could in principle be computed in parallel, given an appropriate initial guess. In this work, we devise a set of procedures that balance the reliability of sequential calculations with the efficiency of parallel computing to various degrees. We demonstrate their usefulness in constructing ground-state PESs of Cr ₂ , Cu ₂ , Mg ₂ , and O ₃ with complete active space self-consistent field and second-order n -electron valence state perturbation theory, and we make recommendations on procedures that maximize the parallelization of PES construction without sacrificing quality. We expect our approach to serve as an important pointer in maximizing the efficiency of PES construction.

Flavin-dependent ’ene’-reductases (ERED) do not naturally rely on light to function. However, EREDs can be repurposed as photoredox catalysts. How light-driven charge transfer (CT) leads to subsequent chemistry remains unclear, particularly regarding the spin dependency of the CT process. Here we computationally investigate whether the CT process is through an intersystem crossing (ISC). We considered four ERED variants and conducted multiscale multireference minimum-energy crossing point (MECP) calculations based on compressed-state multistate pair-density functional theory. We located MECPs between the singlet and triplet manifolds originating from the π → π ^∗ excitation on the flavin. The triplet biradical retains π → π ^∗ character, which according to El-Sayed’s rule implies that ISC is not kinetically favored. None of the variants show much CT character in the triplet state, suggesting that the pathway to produce the CT state is not likely to be spin dependent and is likely through internal conversion.

Multireference calculations are well-suited for describing properties of highly correlated systems. In order to realize the full potential of multireference methods, it is necessary to judiciously choose the active space. It is possible to do this manually, but an automatic procedure for selecting the active space can benefit high-throughput applications and is not susceptible to human error. We previously developed two protocols, GDM-AS and EDM-AS, where the dipole moment calculated at the complete active space self-consistent field (CASSCF) level of theory is used to select the active space, and demonstrated the effectiveness of these protocols when the chosen active spaces are used to find low-lying excitation energies for molecules with non-zero dipole moments. In this work, we demonstrate that using complete active space configuration interaction (CASCI) dipole moments, with the orbitals from the second-order Møller-Plesset perturbation theory, instead of CASSCF dipole moments in our dipole moment active space selection (DM-AS) protocols can significantly improve the efficiency of active space selection compared to our previous methods while maintaining reasonable complete active space second-order perturbation theory (CASPT2) and complete active space pair-density functional theory (CAS-PDFT) vertical excitation energy. We also present new protocols, CASCI-D ² DM-AS and its directional version, designed for charge transfer states. We discuss the performance of our new protocols in their ability to describe excitation energy of not only small organic molecules tested in our previous work, but also larger conjugated polar molecules, nonpolar molecules, and bimolecular complexes, to describe the ground-state spin state of transition metal oxides, and to describe bond dissociation curves of small molecules.

Predicting the optical properties of proteins is critical for the understanding and design of photoactive proteins. Red-shifting flavin-dependent photoenzymes’ spectra beyond the usual blue range h...

Predicting the optical properties of proteins is critical for the understanding and design of photoactive proteins. Red-shifting flavin-dependent photoenzymes’ spectra beyond the usual blue range has become an attractive goal for potential industrial use of photoenzymes for chemical synthesis. However, making such predictions using quantum chemical methods can be computationally intractable, especially if the correct description of the photoexcitation requires multireference methods. We present a sequence-transferrable machine learning framework, LIGand-based PROtein-agnostic SPECTral prediciton (LIG-PROSPECT), to predict the absorption spectra of flavin-dependent photoenzyme variants in the presence of specific substrates. We make use of structural descriptors of cofactor-substrate ensembles as a proxy to represent the 3D conformational features of protein sequences to enable the classification of cyan-light and red-light absorbing peaks for different conformations and the prediction of spectral features of the proteins using linear regression, offering resource-efficient alternatives to multireference quantum mechanical calculations for screening protein sequences for desired photophysical properties.

The ultrafast excited state pathways and dynamics of Ni^II-bipyridine complexes influence the yield of photochemical processes involved in their catalytic cross-coupling reactions. Here we present ultrafast Ni Kβ X-ray emission spectroscopy (XES) and X-ray solution scattering (XSS) of a Ni^II-bipyridine aryl halide complex, [Ni(^t-Bubpy)(o-tol)Br], to quantify the excited state population dynamics and structural changes of the precatalyst. Due to the local spin-sensitivity of XES, the population dynamics of metal-to-ligand charge transfer (MLCT) and metal-centered (MC) excited states is established. A rapid ground state recovery pathway is newly identified, representing a significant deactivation pathway during photocatalysis. Furthermore, the pseudotetrahedral structure of the long-lived MC excited state is unambiguously identified and refined by XSS. The results advance our understanding of the ultrafast relaxation mechanisms that impact the photocatalytic mechanism and yield for Ni^II-bipyridine aryl halide cross-coupling catalysts.

The exploration of potential energy surfaces is central to the understanding of reactions in the ground and excited states. The construction of potential energy surfaces with multireference methods are usually done sequentially, meaning the calculation usually starts from one geometry, and the molecular orbitals and configuration interaction coefficients are carried over to a successive geometry close to the previous geometry, and so on, ensuring fast and reliable convergence to the desired solution. While this is likely to yield smooth and reasonable potential energy surfaces, it can be slow due to having to wait until the previous calculation completes to start the calculation with the next geometry. When multiple computer processors are available, multiple geometries could theoretically be calculated in parallel, given that appropriate initial guess is provided. Based on this, in this work, we devise a set of procedures that balance the reliability of a sequential calculation to obtain the potential energy surfaces with the efficiency enabled by parallel computing to various degrees. We demonstrated the usefulness of the procedures in constructing ground-state potential energy curves of Cr2, Cu2, and Mg2 with complete active space self-consistent field and second-order n-electron valence state perturbation theory. We make a recommendation on procedures which can be used to maximize the parallelization of potential energy surface construction without sacrificing the quality of results.

Selective radical chemistry poses fundamental challenges for modern catalysis. Non-natural photoenzymes, most prominently flavin-dependent "ene"-reductases, have recently emerged as appealing systems to address these challenges by offering unmatched control over chemo-, enantio-, and substrate selectivity, yet their underlying photocatalytic mechanisms remain unclear. Here, we reveal the complete molecular basis of the triple selectivity control in the photoenzymatic radical reactions by the flavin-dependent "ene"-reductase GkOYE-G7 through computational simulations based on multiscale multireference-quantum-mechanics/molecular-mechanics modeling and bias-exchange metadynamics. Our findings demonstrate that control emerges from reaction-level mechanisms rather than binding preferences. We discover that productive photochemistry requires a previously unknown preactivation step involving bond elongation. Stereochemical outcomes likely result from reaction barrier differences, while chemoselectivity, is controlled by crossing points between the ground and excited electronic states around the conical intersection that channel the reaction before competing pathways activate. Substrate scope follows predictable electronic-steric rules, establishing fundamental principles for engineering next-generation photoenzymes with predictable selectivity profiles.