Propagation of Chaos in Contextual Flow Maps

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Chen, Shi, Lin, Zhengjiang, Liu, Kaizhao, Rigollet, Philippe
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910225678729216
author Chen, Shi
Lin, Zhengjiang
Liu, Kaizhao
Rigollet, Philippe
author_facet Chen, Shi
Lin, Zhengjiang
Liu, Kaizhao
Rigollet, Philippe
contents We develop a quantitative statistical theory of transformers in the large-context regime by adopting the abstraction of contextual flow maps (CFMs): dynamical systems that evolve a distinguished token in the presence of a contextual measure across a stack of attention blocks. Within this framework, the finite-context model approximates an idealized infinite-context system in which the contextual measure is replaced by its underlying population, so that the context length $n$ becomes a statistical resource. Exploiting the McKean--Vlasov structure of the dynamics and the classical machinery of propagation of chaos, we establish a forward bound controlling the deviation between the finite- and infinite-context CFMs uniformly along depth, and a backward bound controlling the deviation between the corresponding training trajectories uniformly across iterations of online gradient descent. Both bounds achieve the optimal Wasserstein rate $n^{-1/d}$ for general CFMs and parametric rate $n^{-1/2}$ for a restricted class of CFMs that includes transformers as a special case. The analysis rests on a new Eulerian adjoint formulation of the loss gradient and stability estimates for the resulting forward--adjoint system, both of which may be of independent interest.
format Preprint
id arxiv_https___arxiv_org_abs_2605_16747
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Propagation of Chaos in Contextual Flow Maps
Chen, Shi
Lin, Zhengjiang
Liu, Kaizhao
Rigollet, Philippe
Machine Learning
Analysis of PDEs
Optimization and Control
Probability
Statistics Theory
We develop a quantitative statistical theory of transformers in the large-context regime by adopting the abstraction of contextual flow maps (CFMs): dynamical systems that evolve a distinguished token in the presence of a contextual measure across a stack of attention blocks. Within this framework, the finite-context model approximates an idealized infinite-context system in which the contextual measure is replaced by its underlying population, so that the context length $n$ becomes a statistical resource. Exploiting the McKean--Vlasov structure of the dynamics and the classical machinery of propagation of chaos, we establish a forward bound controlling the deviation between the finite- and infinite-context CFMs uniformly along depth, and a backward bound controlling the deviation between the corresponding training trajectories uniformly across iterations of online gradient descent. Both bounds achieve the optimal Wasserstein rate $n^{-1/d}$ for general CFMs and parametric rate $n^{-1/2}$ for a restricted class of CFMs that includes transformers as a special case. The analysis rests on a new Eulerian adjoint formulation of the loss gradient and stability estimates for the resulting forward--adjoint system, both of which may be of independent interest.
title Propagation of Chaos in Contextual Flow Maps
topic Machine Learning
Analysis of PDEs
Optimization and Control
Probability
Statistics Theory
url https://arxiv.org/abs/2605.16747