{"id":"arxiv/2608.25055","url":"https://arxiv.org/abs/2608.25055","tags":["math.AP"],"score":0,"title":"A Mean-Field Theory of Transformers: Well-Posedness of the Coupled Data--Parameter Dynamics and Global Convergence of Training","domain":"math.AP","source":"arxiv","authors":["Michael Herty","Hailiang Liu"],"reviews":0,"subject":"math.AP","summary":"We develop a rigorous mean-field theory for transformer networks that captures two large-scale limits inherent in the architecture: the number of tokens $N\\to\\infty$ in the input sequence and the number of attention heads $H\\to\\infty$ in each layer. It further considers an infinite number of layers leading to a time-continuous formulation. The resulting framework couples two interacting mean-field objects: a token distribution $μ_t\\in\\mathcal P(\\mathbb R^d)$, which evolves through network depth $t\\in[0,T]$ according to a McKean--Vlasov transport equation, and the attention-parameter…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-08-25","reactions":{}}