Legeza, Örs (2025.11.01 - 2026.04.30)

Abstract: The numerical simulation of quantum systems characterized by strong interactions between atomic spins or mobile electrons, beyond the scope of perturbation methods, is a fundamental aspect of contemporary physics. However, this presents a significant challenge due to the exponential scaling of computational resources with system size. Consequently, the development of algorithms capable of reducing this complexity to a polynomial form constitutes one of the most actively researched domains within the scientific community today.

The density matrix renormalization group (DMRG) algorithm is one such method that offers an added advantage. It divides the relevant tensor algebra into millions of independent subtasks based on the remaining quantum numbers. This makes it an ideal environment for MPI and GPU-based massive parallelization. In 2021-2024, we have already conducted simulations on various quantum systems using our hybrid CPU and multiple GPU-accelerated applications. As a result, we have published five articles and a handbook in Phys Rev B, J. Chem Theory and Comput, and arXiv.

In the proposed research program, we aim to investigate systems containing novel metal atoms that are inaccessible through conventional quantum chemistry methods. Furthermore, we intend to precisely determine the electronic structure of intermediate states generated during novel catalytic reactions. Additionally, we plan to conduct simulations of the static and dynamic properties of strongly correlated two-dimensional fermion lattice models.

Requirements: To conduct the tests, we require a dedicated node equipped with a local disk. During the test, access to all eight GPU cards is essential, and only our application should be executed. Failure to adhere to these requirements will result in unreliable test results. Similarly, our program utilizes the full system capacity of the GPU node. Consequently, no other jobs can be executed simultaneously to avoid conflicts in memory management and GPU utilization.