Paperlane

no. 12 · Tuesday 10 November 2026

A Geometric Theory for Nonequilibrium Transformer Dynamics

Aythami Bethencourt de León and Darryl D. Holm

math.MP · via cs.LG · 29 pp

Abstract

We develop a variational and geometric theory based on statistical physics for the joint evolution of Transformer attention and token representations at fixed model parameters. To this end, we introduce a free energy for the attention and model the evolution of the attention weights by its Fisher–Rao gradient flow. At fixed token representations, we prove that the softmax matrix is the unique free-energy minimiser and that every positive attention trajectory converges to it. For prescribed time-dependent representations, we show that the attention dynamics becomes linear after a change to logit variables. This result yields tracking bounds relative to the instantaneous softmax and sufficient conditions for attention to converge to a fixed matrix. Hamiltonian dynamics for the representations and their momenta are then coupled to the attention relaxation driven by the same energy. This coupling results in an energy-dissipation law. We also develop two alternative Hamiltonians incorporating the attention value map. We prove global well-posedness of the attention flow and all three coupled systems.

comments
29 pages Submitted to Proc Roy Soc A
fields
Mathematical Physics, Probability
pdf ↗ page arXiv:2611.10691 ↗