Memory-Conditioned Flow-Matching for Stable Autoregressive PDE Rollouts
Victor Armegioiu · 2026
Read Memory-Conditioned Flow-Matching for Stable Autoregressive PDE Rollouts by Victor Armegioiu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-06689.txt.
Description
Victor Armegioiu tackles a weakness of generative solvers for partial differential equations: they predict one step well but drift badly over long rollouts. The drift is worst when each step must rebuild fine details that a coarse grid leaves out. Using a classical projection formalism, the paper shows that discarding unresolved variables leaves an exact evolution law with a memory term that memoryless solvers simply ignore.
The proposed fix gives the generator a compact memory state, updated online and fed into its denoising step, which shapes how missing fine scales are filled in. The author proves stability of the resulting process and derives rollout error bounds that separate memory mistakes from generation mistakes. On compressible flows with shocks and multiscale mixing, the method stayed accurate and far steadier over long horizons.
Catalog record
| CREATOR | Victor Armegioiu |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 115 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-06689