Universal approximations of quasilinear PDEs by finite distinguishable particle systems
Thierry Paul, Emmanuel Trélat
TL;DR
The paper establishes a constructive link between quasilinear PDEs and finite particle systems by showing that sufficiently regular PDE solutions can be approximated, with error decaying like $1/\log N$, by carefully designed $N$-particle dynamics derived from mollified operators. The method proceeds in two steps: first approximate the PDE operator with a bounded, mollified operator $A_\varepsilon$ and obtain a correspondingly mollified solution $y_\varepsilon$, then discretize via tagged partitions to build a particle system $y_\varepsilon^N$ that converges to $y_\varepsilon$ as $N\to\infty$; combining these steps yields quantitative convergence rates for $y_\varepsilon^N$ to $y$. The authors provide explicit particle constructions for a broad class of quasilinear PDEs, derive convergence in $L^2$ (and in $L^\infty$ under suitable embeddings), and present concrete examples (transport, Burgers, KdV). They also connect the analysis to graph-limit ideas and discuss extensions such as the backward heat equation and variational inequalities, offering a versatile framework for linking continuum PDEs to finite, interpretable particle systems with rigorous error control.
Abstract
In this paper, we prove that sufficiently regular solutions of any quasilinear PDE can be approximated by solutions of systems of N distinguishable particles, to within 1/ ln(N ). This intruiguing result is related to recent developments in graph limit theory.
