Structural Origin of Water Heat Capacity Anomaly from Classical and Quantum Simulations

What the math gives to ML

The paper identifies a quantitative two-state mechanism: anomalous heat capacity is generated by temperature-dependent fluctuations between low-density-like and high-density-like local structures, while quantum effects mainly suppress high-frequency vibrational contributions. The transferable asset is the separation of fast fluctuations from a slow, latent structural coordinate and the relation between susceptibility of that coordinate and excess variance in an observable. A neural analogue is a regime-switching module whose soft gate is trained as a calibrated two-state variable, with a susceptibility monitor that detects transition regions and allocates model capacity there. This is most promising for world models, sequence models, and mixture-of-experts systems with genuinely distinct dynamical regimes.