Maximum Hidden Irreversibility Reveals Latent Rotational Dynamics and Predicts Emergent Grokking in Neural Networks

This is Version 2 of the paper. Compared to Version 1, this release includes: Reference-quality and numerically stable implementation of Maximum Hidden Irreversibility (MHI)- Mixture-of-Experts (MoE) extension with routing entropy- Analysis of rotational (antisymmetric) dynamics in neural representations- Improved toy model simulations for demonstration and testing- Updated experiments, figures, and clarifications throughout the paper Note: results presented are proof-of-concept simulations illustrating theoretical predictions. Empirical validation on real training runs is in preparation.