# Viscous conservation laws

> Momentum breaks the conservation laws of neural network training, and friction decides by how much. Exact drift laws, proved in Lean 4.

October 2026. Section: AI, LLMs, ML and cognitive neuroscience.

Paper: [Friction keeps the conservation laws of training alive](https://doi.org/10.5281/zenodo.23510521)

## Story

On 6 October 2026 I asked whether gradient descent being the slow limit of motion through a thick fluid leads anywhere new. It led to exact drift laws under momentum, and the paper credits each piece that turned out to be known already.

## Links

- [Project](https://github.com/pragyaangaur/Viscous-Conservation-Laws)

## See also

- [The melting temperature of a next-token distribution](https://pragyaangaur.dev/work/token/index.md)
- [Magnus turning invariants](https://pragyaangaur.dev/work/magnus/index.md)
- [Cauchy-Schwarz for real powers](https://pragyaangaur.dev/work/cauchy/index.md)

By [Pragyaan Gaur](https://pragyaangaur.dev/). The full index is at https://pragyaangaur.dev/llms.txt and https://pragyaangaur.dev/index.json.
