Function graph transformers use graph measures to provide a measure-theoretic framework where standard transformer components universally approximate operators between function spaces while preserving single-valued function outputs.
Mean field, hydrodynamic and graph limits for deterministic interacting particle systems: a survey with quantitative estimates
5 Pith papers cite this work. Polarity classification is still indexing.
5
Pith papers citing it
abstract
We present a unified framework, with quantitative estimates, for deterministic interacting particle systems whose pairwise interactions may depend on heterogeneous labels. Heterogeneity is kept at every level by adding a frozen label variable $x\in\Omega$ to the state. Within this framework we compare several limiting procedures: the direct continuum / graph limit, the mean field limit yielding a Vlasov equation on the extended space of labels and states, the Liouville lift of the particle system together with propagation of chaos through marginals of arbitrary order, and the hydrodynamic moment closures. We give a common language for these limits and identify precisely where the various passages commute and where they do not; in particular, we separate the continuum / graph limit equation from the classical hydrodynamic Euler equations and characterize when the former arises as a moment closure of the latter (linearity in $(\xi,\xi')$ or monokinetic ansatz). Along the way, we prove quantitative convergence estimates for the graph limit and for the passages from particles or Liouville to Vlasov, and we discuss the limitations of the framework, in particular concerning singular kernels and stochastic dynamics. The paper is written as a survey with original contributions, with an emphasis on estimates, examples, and a clear delineation of scope.
years
2026 5representative citing papers
Proves O((α_n n)^{-1/2}) convergence of GNDE trajectories and adjoints to Graphon-NDE limits on sparse random graphs, with DTO/OTD consistency and experimental support for zero-shot transfer.
A measure-based nudging framework assimilates smoothed macroscopic observations into microscopic mean-field particle dynamics using Wasserstein transport velocities on probability measures.
Global well-posedness and quantitative flocking are shown for Lagrangian p-alignment dynamics; Eulerian variables are constructed via pushforward and disintegration, with defect terms vanishing asymptotically under heavy-tailed kernels to give mono-kinetic closure and mean-field convergence.
An abstract theorem bounds the error between exact solutions of quasilinear PDEs and their finite-dimensional approximations obtained via regularization followed by sampling-reconstruction, with explicit trade-offs between scales and an example using boundary-preserving mollifiers.
citing papers explorer
-
Function graph transformers universally approximate operators between function spaces
Function graph transformers use graph measures to provide a measure-theoretic framework where standard transformer components universally approximate operators between function spaces while preserving single-valued function outputs.
-
Zero-Shot Size Transfer for Neural ODEs on Sparse Random Graphs: Graphon Limits and Adjoint Convergence
Proves O((α_n n)^{-1/2}) convergence of GNDE trajectories and adjoints to Graphon-NDE limits on sparse random graphs, with DTO/OTD consistency and experimental support for zero-shot transfer.
-
Multiscale Nudging: From Macroscopic Observations to Microscopic Dynamics
A measure-based nudging framework assimilates smoothed macroscopic observations into microscopic mean-field particle dynamics using Wasserstein transport velocities on probability measures.
-
Lagrangian formulation and Eulerian closure in alignment dynamics
Global well-posedness and quantitative flocking are shown for Lagrangian p-alignment dynamics; Eulerian variables are constructed via pushforward and disintegration, with defect terms vanishing asymptotically under heavy-tailed kernels to give mono-kinetic closure and mean-field convergence.
-
Boundary-compatible interacting approximations of quasilinear PDEs on bounded domains
An abstract theorem bounds the error between exact solutions of quasilinear PDEs and their finite-dimensional approximations obtained via regularization followed by sampling-reconstruction, with explicit trade-offs between scales and an example using boundary-preserving mollifiers.