Coupled cluster methods for local and ultrafast spectroscopy using graphical processing units

We have made significant progress with theory and computer implementations in this reporting period. In particular, we have in collaboration with the group at Stanford University we have developed a low-rank full configuration interaction implementation that will make possible much larger active spaces than using standard approaches. In the next reporting period, we expect to employ this code for production benchmarks. We have also completed a formulation and implementation of coupled cluster theory for excited state conical intersections. These results will shortly be submitted for publications. Another development is the implementation of a novel algorithm for Cholesky decomposition of the two-electron integrals that enter the electronic structure equations - the new approach can be applied to very large systems and we have demonstrated the use for molecules with several thousand atoms. We have also developed novel techniques for performing multi-level Hartree-Fock calculations. These optimization techniques will significantly improve the convergence of the working equations that need to be solved. Finally, we have completed the implementation of the complex coupled cluster code - this new code will be used to study time-dependent electron dynamics. We have completed a new implementation of the CC3 method that will be used in several collaboration projects with experimental groups. These projects mainly focus of X-ray experiments that are carried out at large scale facilities. The time-dependent coupled cluster project has now matured and this will now enter the phase of optimization of the methodology and we will focus on charge migration in electronic excited states. Implementation on advanced computer architectures is also progressing according to plan, however, the efficiency of the algorithms is not completely satisfactory and will require additional work. The CCGPU project is now close to the end and all the major milestones have been completed - both with respect to the GPU development and time-dependent coupled cluster. We have recently developed a GPU implementation of CC3 and the initial tests look promising.

The CCGPU project has been essential for laying the groundwork for several recent successful applications - this includes among others a Marie Curie ITN network and an ERC Advanced grant. Many of the scientific works produced in this project are well cited and will be important for future developments in the field.

In this project, we will enable the use of coupled cluster (CC) theory for the description of excited state nuclear dynamics. Long standing claims regarding the deficiency of CC theory to describe regions around conical intersections on potential energy surfaces and thus its inability to describe excited state dynamics for photoactive systems, will be revised. In addition, we have formulated a theory that enables CC methods for excited states to be written as a Hermitian eigenvalue problem which always has well defined solutions. Thus, the proposed project will enable the use of the accurate CC methods for dynamics calculations. These developments will be carried out in a multi-level framework to allow for the accurate description of systems and properties in a computationally efficient way. Furthermore, implementations on graphics processing units (GPUs) will be utilized to achieve a short time-to-solution. The proposed project will therefore not only encompass game-changing theoretical developments, but also build up competence in GPU-enabled quantum chemistry in Norway.