Quantum many-body systems in and out of equilibrium
||Quantum many-body systems in and out of equilibrium|
||SNIC Medium Compute|
||Hugo Strand <Hugo.Strand@oru.se>|
||2021-02-02 – 2021-09-01|
Understanding the dynamics of quantum many-body systems is central in multiple fields of physics. Both for the low temperature mesoscopic systems and cold atom setups used to build quantum computers and the emerging spectroscopies using neutrons, pump probe setups, and free electron laser sources.
This project will perform classical computer simulations of quantum many-body systems and their dynamical response, both in equilibrium and the real-time dynamics out of equilibrium. The calculations will be performed using a number of open source community codes that we are developing.
For equilibrium single-particle and two-particle dynamics we will use the Toolbox for Research on Interacting Quantum Systems (TRIQS) and its Continuous Time Hybridization-expansion (CTHYB) quantum Monte Carlo solver. CTHYB scales linearly over independent Markov chains to thousands of compute nodes, tested on, e.g., Piz Daint (Cray XC40) at the Swiss National Computing Centre. For lattice spectroscopies the Two Particle Response Function toolbox (TPRF), with hybrid MPI+OpenMP parallelization, will be used. TRIQS, CTHYB and TPRF are open sourced community projects and developed in collaboration with the Simons Foundation's Center for Computational Quantum-Physics, New York, USA.
For studying real-time dynamics we will use the Non-Equilibrium Systems Simulation package (NESSi), MPI parallelized over k-points, and the Pseudo-Particle Strong Coupling solver (PPSC), with OpenMP parallelization over diagram integrals and MPI parallelization over diagrams. NESSi and PPSC is developed in collaboration with groups at the University of Fribourg, Switzerland, and the Jožef Stefan Institute, Slovenia.
Apart from these established community codes the project will also explore new numerical methods for quantum many-body systems. In collaboration with groups at the department of Physics and department of Chemistry at the University of Michigan, USA, we will explore applications of spectral methods, starting with Legendre spectral methods for the imaginary-time Dyson equation. Currently we are investigating the scaling of an implicit linear solver formulation using a sparse sampling preconditioner. Separately in collaboration with a group at University of Hamburg, Germany we will explore novel stochastic algorithms for real-time dynamics.
One central component of the development of these codes will -- in the light of recent announcements from EuroHPC JU -- be to port the central algorithms to GPU accelerator platforms, in order to be able to benefit of coming large-scale GPU-based HPC systems such as Lumi at CSC in Finland.