New Physics-Based Brain Model Uses Hamiltonian Dynamics to Improve BCI Decoders

A new paper on arXiv (2607.10439) by Dibakar Sigdel introduces a physics-inspired approach to modeling brain dynamics by treating the motor cortex as a port-Hamiltonian system (pHS). In this framework, neural activity is decomposed into a conservative interconnection (gyroscopic coupling between neural phasors) and a dissipative port driven by a graph neural network (GNN) surrogate that models power-law energy decay. A metriplectic integrator evolves the phasor state, ensuring structure preservation, while a Fluctuation-Dissipation-consistent noise channel adds stochasticity at body temperature. Training on real EEG cycles from the PhysioNet EEGMMIDB dataset (with 3 held-out subjects) yields strong predictive performance and passes three key rungs of scale-free criticality: near-critical branching ratio (σ≈1), a 1/f power-law spectrum, and long-range detrended fluctuation analysis (DFA) correlations.

The model's ability to generate closed-loop neuromodulation signals that restore phase-locking in desynchronized inputs suggests a direct path toward improved brain-computer interface (BCI) decoders. By preserving the underlying Hamiltonian structure of neural dynamics, this approach outperforms black-box models and offers better interpretability and robustness. Sigdel's work highlights how principled physical modeling can bridge the gap between theoretical neuroscience and practical neurotechnology, potentially enabling more natural and adaptive BCI systems for motor rehabilitation and neural control.