Conservative Discrete Structure Stabilizes Autoregressive Rollouts in a 1D Drift Diffusion Poisson Benchmark

Yufeng Wang*, Lu Wei*, Haibin Ling † {\dagger} · 2026

Read Conservative Discrete Structure Stabilizes Autoregressive Rollouts in a 1D Drift Diffusion Poisson Benchmark by Yufeng Wang*, Lu Wei*, Haibin Ling † {\dagger} (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-2606-01366.txt.

Description

Wang, Wei, and Ling test what keeps learned plasma rollouts stable. Learned surrogates can match short-term states yet fail over long rollouts. Charge accounting and density limits are often not enforced. Field reconstruction may also clash with the Poisson equation. Their test bed is a 1D drift diffusion Poisson benchmark. Walls have fixed potential and zero species flux. The benchmark checks conservation, not full sheath physics.

A conservative finite volume core alone nearly nails the rollout. Its error sits near roundoff level in their headline test. It beats direct learned regressors by many orders of magnitude. Across 64 settings, it wins rollout error in 60 cases. Yet it wins one-step error in just 19 cases. Local conservative structure matters more than one-step accuracy.

Catalog record

CREATORYufeng Wang*, Lu Wei*, Haibin Ling † {\dagger}
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE71 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2606.01366

Browse the full library · About the Tower of Babel