Quantum-inspired framework for computational fluid dynamics

Abstract

Computational fluid dynamics is both a thriving research field and a key tool for advanced industry applications. However, the simulation of turbulent flows in complex geometries is a compute-power intensive task due to the vast vector dimensions required by discretized meshes. We present a complete and self-consistent full-stack method to solve incompressible fluids with memory and run time scaling logarithmically in the mesh size. Our framework is based on matrix-product states, a compressed representation of quantum states. It is complete in that it solves for flows around immersed objects of arbitrary geometries, with non-trivial boundary conditions, and self-consistent in that it can retrieve the solution directly from the compressed encoding, i.e. without passing through the expensive dense-vector representation. This framework lays the foundation for a generation of more efficient solvers of real-life fluid problems. Simulating turbulent fluids is a major computational challenge, the main obstacle being the large size of discretized meshes required to accurately describe turbulent flows. The authors develop a quantum-inspired framework, based on matrix product states, to solve for flows around immersed bodies with complexity scaling logarithmically in the mesh size.