# FlowForge: A Staged Local Rollout Engine for Flow-Field Prediction

Source: [FlowForge: A Staged Local Rollout Engine for Flow-Field Prediction](<https://www.sota2.com/research/paper/69e85bd97d0172140ec81179>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-04-28T05:00:12.890Z

Authors: Xiaowen Zhang, Ziming Zhou, Fengnian Zhao, David L. S. Hung

Published: 2026-04-21T01:04:26.000Z

## Abstract

Deep learning surrogates for CFD flow-field prediction often rely on large, complex models, which can be slow and fragile when data are noisy or incomplete. We introduce FlowForge, a staged local rollout engine that predicts future flow fields by compiling a locality-preserving update schedule and executing it with a shared lightweight local predictor. Rather than producing the next frame in a single global pass, FlowForge rewrites spatial sites stage by stage so that each update conditions only on bounded local context exposed by earlier stages. This compile-execute design aligns inference with short-range physical dependence, keeps latency predictable, and limits error amplification from global mixing. Across PDEBench, CFDBench, and BubbleML, FlowForge matches or improves upon strong baselines in pointwise accuracy, delivers consistently better robustness to noise and missing observations, and maintains stable multi-step rollout behavior while reducing per-step latency.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Flow Field Prediction on CFDBench Tube](<https://www.sota2.com/research/sota/flow-field-prediction-on-cfdbench-tube.md>): Flow Field Prediction; CFDBench Tube; RMSE: 0.1035.

- [Flow Field Prediction on CFDBench Cylinder](<https://www.sota2.com/research/sota/flow-field-prediction-on-cfdbench-cylinder.md>): Flow Field Prediction; CFDBench Cylinder; RMSE: 0.0258.

- [Flow Field Prediction on CFDBench Dam](<https://www.sota2.com/research/sota/flow-field-prediction-on-cfdbench-dam.md>): Flow Field Prediction; CFDBench Dam; RMSE: 0.0222.

- [Flow Field Prediction on PDEBench Diff-React](<https://www.sota2.com/research/sota/flow-field-prediction-on-pdebench-diff-react.md>): Flow Field Prediction; PDEBench Diff-React; RMSE: 0.0837.

- [Flow Field Prediction on PDEBench Rand-M0.1](<https://www.sota2.com/research/sota/flow-field-prediction-on-pdebench-rand-m0-1.md>): Flow Field Prediction; PDEBench Rand-M0.1; RMSE: 0.0451.

- [Flow Field Prediction on BubbleML FB-Gravity](<https://www.sota2.com/research/sota/flow-field-prediction-on-bubbleml-fb-gravity.md>): Flow Field Prediction; BubbleML FB-Gravity; RMSE: 0.1869.

- [Flow Field Prediction on BubbleML PB-Subcooled](<https://www.sota2.com/research/sota/flow-field-prediction-on-bubbleml-pb-subcooled.md>): Flow Field Prediction; BubbleML PB-Subcooled; RMSE: 0.3798.

- [Flow Field Prediction on BubbleML FB-VelScale](<https://www.sota2.com/research/sota/flow-field-prediction-on-bubbleml-fb-velscale.md>): Flow Field Prediction; BubbleML FB-VelScale; RMSE: 0.5382.

- [Flow Field Prediction on PDEBench Rand-M1.0](<https://www.sota2.com/research/sota/flow-field-prediction-on-pdebench-rand-m1-0.md>): Flow Field Prediction; PDEBench Rand-M1.0; RMSE: 0.1351.

- [Flow Field Prediction on BubbleML PB-Gravity](<https://www.sota2.com/research/sota/flow-field-prediction-on-bubbleml-pb-gravity.md>): Flow Field Prediction; BubbleML PB-Gravity; RMSE: 0.2898.

- [Flow Field Prediction on CFDBench Cavity](<https://www.sota2.com/research/sota/flow-field-prediction-on-cfdbench-cavity.md>): Flow Field Prediction; CFDBench Cavity; RMSE: 0.2949.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2604.18953>)

[PDF](<https://arxiv.org/pdf/2604.18953.pdf>)