A Data-Free Physics-Informed Neural Operator for Level-Set Interface Advection
Not specified in the provided content
Abstract
This paper presents a data-free physics-informed neural operator for level-set interface advection that maps an initial interface to its full spatiotemporal trajectory without using reference solutions.
Reality Card
The data-free operator achieves a relative L2 error of 1.614% on a reversed single vortex, outperforming the supervised baseline by a factor of 4.4.
The data-free operator conserves enclosed area 2.7 times better than the supervised baseline despite a larger field error.
The performance may vary significantly based on the validity of the eikonal constraint, which affects the accuracy of the operator under different conditions.
Paper to code
Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.