Flowers: A Warp Drive for Neural PDE Solvers
Till Muser, Alexandra Spitzer, Matti Lassas, Maarten V. de Hoop, Ivan Dokmanić · 2026
Read Flowers: A Warp Drive for Neural PDE Solvers by Till Muser, Alexandra Spitzer, Matti Lassas, Maarten V. de Hoop, Ivan Dokmanić (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-2603-04430.txt.
Description
Muser and colleagues solve PDEs with warps alone in Flowers. Only pointwise channel mixing and a multiscale scaffold join the warps. No Fourier multipliers, dot product attention, or convolutional mixing appear. Each head warps mixed features by a predicted displacement field, sampled sparsely at source points. Stacked multiscale residual warps give global adaptive interactions at linear cost. Theory ties the design to conservation flow maps, waves in inhomogeneous media, and a kinetic limit. A 17M model beats similar baselines, and a 150M model tops recent transformer foundation models.
Catalog record
| CREATOR | Till Muser, Alexandra Spitzer, Matti Lassas, Maarten V. de Hoop, Ivan Dokmanić |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 71579 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.04430