Periodic Reporting for period 1 - CC4DYN (A quantum chemical approach to dynamic properties of real materials)
Reporting period: 2023-08-01 to 2026-01-31
Computational materials science, leveraging ab initio simulations and high-performance computing, is poised to play a pivotal role in realising the vision of "materials by design". However, the goal of discovering game-changing materials with significant scientific and industrial relevance demands highly accurate ab initio methods capable of addressing both excited state and ground state properties of atoms, molecules, and solids. To date, the computational complexity of these methods has meant that approaches with systematically improvable accuracy for condensed matter systems—such as coupled-cluster (CC) theories—are primarily confined to studying ground state properties within the clamped-nuclei approximation.
This project aims to induce a paradigm shift in how we study the vibrational and optical properties of real materials by introducing a series of novel methodologies. On one hand, we propose to reduce the computational cost of time-dependent equation-of-motion coupled-cluster (TD-EOM-CC) theory compared to existing approaches, making it feasible to study larger and more complex systems. On the other hand, we aim to implement coupled-cluster atomic forces and combine them with the framework of machine-learning force fields.
Together, these proposed methods have the potential to achieve an unprecedented level of accuracy and scalability for the prediction of a wide range of material properties, including optical spectra and phonon frequencies. By leveraging these new approaches, we aim to resolve several long-standing discrepancies between theoretical predictions and experimental data, particularly for the dynamic properties of defects, molecular crystals, and layered materials.
This project aims to induce a paradigm shift in how we study the vibrational and optical properties of real materials by introducing a series of novel methodologies. On one hand, we propose to reduce the computational cost of time-dependent equation-of-motion coupled-cluster (TD-EOM-CC) theory compared to existing approaches, making it feasible to study larger and more complex systems. On the other hand, we aim to implement coupled-cluster atomic forces and combine them with the framework of machine-learning force fields.
Together, these proposed methods have the potential to achieve an unprecedented level of accuracy and scalability for the prediction of a wide range of material properties, including optical spectra and phonon frequencies. By leveraging these new approaches, we aim to resolve several long-standing discrepancies between theoretical predictions and experimental data, particularly for the dynamic properties of defects, molecular crystals, and layered materials.
The project is progressing well in the first half of the funding period. A functional prototype of TD-EOM-CC for solids has been implemented and tested, with current efforts focused on optimising its computational efficiency. Preliminary results for hexagonal BN phases have been obtained, but the main goal remains refining the method for broader applications. Low-rank tensor approximations are also being explored to reduce computational complexity, with full development expected in the near future. Progress is also being made in implementing nuclear gradients within Coupled-Cluster (CC) theory, which are essential for training machine-learned force fields (MLFFs) for complex materials. While a prototype is already operational for small test systems, further optimization is needed for large-scale use. MLFFs are being trained with CC potential energies, and their integration with nuclear gradients is expected soon. In addition, the study of Th-doped CaF2 crystals is progressing, with new insights into the atomic structure of Th defects providing a foundation for upcoming TD-EOM-CC calculations. Preliminary tests on the optical properties of hexagonal boron nitride and data for solid hydrogen phases are helping refine the methods for simulating complex systems and informing ongoing developments in the project.
The project has made significant strides, with several noteworthy achievements. A recent publication in Nature Communications investigated discrepancies in noncovalent interaction energies for large molecules, providing fresh insights into longstanding issues between two prominent electronic structure theories. Additionally, some of our work introduces the first application of machine-learned force fields trained on periodic coupled-cluster theory results for covalent crystals, paving the way for future research in this area.
In the field of catalysis, we successfully demonstrated the accurate prediction of CO adsorption on Pt(111) using periodic coupled-cluster theory. This work confirms that highly accurate adsorption energies and correct adsorption sites can be determined, and sets the stage for future studies on reaction mechanisms with exceptional accuracy.
Finally, we have made ongoing improvements to our open-source simulation software, Cc4s, incorporating the novel methods developed during the project. Gaining increasing attention in the materials science community, Cc4s is now being applied to a wide range of problems, from molecular adsorption and phase diagrams to the study of defects.
In the field of catalysis, we successfully demonstrated the accurate prediction of CO adsorption on Pt(111) using periodic coupled-cluster theory. This work confirms that highly accurate adsorption energies and correct adsorption sites can be determined, and sets the stage for future studies on reaction mechanisms with exceptional accuracy.
Finally, we have made ongoing improvements to our open-source simulation software, Cc4s, incorporating the novel methods developed during the project. Gaining increasing attention in the materials science community, Cc4s is now being applied to a wide range of problems, from molecular adsorption and phase diagrams to the study of defects.