Smooth particle methods without smoothing
This work addresses the problem of accurate and stable particle transport in computational fluid dynamics, offering a significant improvement over standard particle methods by eliminating the need for frequent remappings.
This paper introduces a new class of particle methods with deformable shapes that converge in the uniform norm without requiring remappings, extended overlapping, or vanishing moments. The methods achieve high accuracy and robustness, with quadratically-transformed particles transported without remappings over long periods.
We present a new class of particle methods with deformable shapes that converge in the uniform norm without requiring remappings, extended overlapping or vanishing moments for the particles. The crux of the method is to use polynomial expansions of the backward characteristic flow to transport the numerical particles with improved accuracy: in the first order case the method consists of representing the transported density with linearly-transformed particles, the second order version computes quadratically-transformed particles, and so on. For programming purposes we provide explicit implementations of the resulting LTP and QTP schemes that only involve pointwise evaluations of the forward characteristic flow, and also come with local indicators for the accuracy of the corresponding transport scheme. Numerical tests using different transport problems demonstrate the accuracy of the proposed methods compared to standard particle schemes, and establish their robustness with respect to the remapping period. In particular, it is shown that QTP particles can be transported without remappings on very long periods of times, without hampering the accuracy of the numerical solutions. Finally, a dynamic criterion is proposed to automatically select the time steps where the particles should be remapped. The strategy does not require additional inter-particle communications, and it is validated by numerical experiments.