Methodological developments in electronic structure theory and chemical dynamics

Electronic structure methods are vital, yet they are often too computationally expensive. Here, the authors develop machine learned density matrices to fully represent electronic structures in a computationally cheap and accurate way.

In machine learning of molecular properties, adequate molecular representation is crucial, as minor structural changes result in significant differences in the activity of interest. Here, the authors use a lower-dimensional representation of a molecular manifold to embed electronic attributes and retain the chemical intuition of molecular interactions, invariant to translational and rational degrees of freedom of the molecular surface, to ease model complexities and facilitate generalization even with relatively small datasets.

The precise definition of the surface of atoms and molecules has long been an open issue. Here, the authors combine theory and experimental thermodynamic measurements to define atomic and molecular surfaces from electron iso-density surfaces contoured at a value of 0.0016 a.u.

Normal mode analysis is a crucial step in structural biology, but is based on an expensive diagonalisation of the system’s Hessian. Here the authors present INCHING, a GPU-based approach to accelerate this task up to >250 times over current methods for macromolecular assemblies.

Ab-initio methods often require many steps before the simulated structure can escape trivial local energy minimums. Here, authors show that tiered tensor transform (3T) can be used to explore complex chemical reactions with modest count of DFT steps.

A balanced description of static and dynamic correlation effects has long been a challenge for most density functional methods. Here, the authors proposed a new hybrid multi configuration density functional theory method that is shown to give satisfactory results for general purpose.

Molecular dynamics is a common tool to study microscopic physicochemical systems, however, it is limited by the inhability to form and break chemical bonds. Here the authors present a method to modify traditional force-fields implementing bond dissociation and bond forming.

Transition states are central to chemical reactions. Here, the authors derive the analytical Hessians from a neural network potential for organic reactions and yield more efficient and robust saddle point optimization than using quasi-Newton updates.

Electronic simulations of large systems are computationally demanding. Here, authors develop DeePTB, a deep learning approach for efficient tight-binding calculations with ab initio accuracy, enabling million-atom simulations at finite temperatures.

Hybrid density functionals are crucial for accurate materials calculations, yet their application is limited by the computational cost. Here, authors overcome this efficiency bottleneck through deep learning, enabling large-scale hybrid density functional calculations.

Calculating electronic spectra of large systems is computationally challenging. Here, the authors combine exact short-time dynamics with approximate frequency space methods to capture narrow features embedded in a dense manifold of smaller peaks.

The extent of problems in quantum chemistry for which quantum algorithms could provide a speedup is still unclear, as well as the kind of speedup one should expect. Here, the authors look at the problem of ground state energy estimation, and gather theoretical and numerical evidence for the fact that an exponential quantum advantage is unlikely for generic problems of interest.

Quantum computing offers a promising approach to solving electronic-structure problems, but a quantitative description of chemical systems while minimizing computing resources is an essential challenge. Here, the authors provide a shortcut towards chemically accurate quantum computations by approaching the complete-basis-set limit through coupling the density-based basis-set corrections approach, applied to any given variational ansatz, to an on-the-fly crafting of basis sets specifically adapted to a given system and user-defined qubit budget.