Machine Learning Enhanced Collision Operator for the Lattice Boltzmann Method Based on Invariant Networks

Mario Christopher Bedrunka, Tobias Horstmann, Ben Picard, Dirk Reith, Holger Foysi · 2024

Read Machine Learning Enhanced Collision Operator for the Lattice Boltzmann Method Based on Invariant Networks by Mario Christopher Bedrunka, Tobias Horstmann, Ben Picard, Dirk Reith, Holger Foysi (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-08229.txt.

Description

Mario Christopher Bedrunka and co-authors build a neural collision operator for a lattice-based fluid simulation method, in a paper from 2024. The collision step is where machine learning can most easily enter this classic solver. A good operator must respect the symmetries of fluid motion.

Their network is built to be invariant, acting on an operator that already respects symmetry. It tunes the relaxation rates of the non-physical moments in the simulation. Training used forced turbulence runs with a known energy spectrum. The loss pushed the cheap, coarse simulation to match the spectrum of a full one, with extra control of damping at fine scales. On a standard vortex test, the operator predicted the flow well even at very coarse resolution.

Catalog record

CREATORMario Christopher Bedrunka, Tobias Horstmann, Ben Picard, Dirk Reith, Holger Foysi
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE62 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-08229

Browse the full library · About the Tower of Babel