FieldSeer I: Physics-Guided World Models for Long-Horizon Electromagnetic Dynamics under Partial Observability

arXiv 25.12 2025 61.5 method, benchmark, application

TLDR

Geometry-aware world model forecasts electromagnetic field dynamics from partial observations, enabling interactive digital twins for photonic design.

Reasoning

The paper introduces a novel geometry-conditioned world model for electromagnetic dynamics, demonstrating strong performance on simulated FDTD benchmarks and enabling interactive edits. However, it lacks real-world experimental validation and is limited to 2-D TE waveguides.

Read-first score

Read-first score 61.5, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 45.

Recency 8%

86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Methodology quality 25%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=baseline,benchmark,metric,result

Topical relevance 42%

64.3

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 175.

Keyword Scores

world model

world dynamics prediction

interactive world model

world simulator

generative world model

video world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Geometry-aware world model for forecasting electromagnetic field dynamics from partial observations in 2-D TE waveguides
Closed-loop rollouts conditioned on scalar source action and structure/material map
Training in symmetric-log domain for numerical stability
Edit-after-prefix geometry modifications without re-assimilation

Methodology

FieldSeer I is a geometry-aware world model that assimilates a short prefix of observed fields, conditions on a scalar source action and structure/material map, and generates closed-loop rollouts in the physical domain. Training is performed in a symmetric-log domain to ensure numerical stability. The model is evaluated on a reproducible FDTD benchmark with 200 unique simulations using a structure-wise split, comparing against GRU and deterministic baselines across three practical settings.

Key Results

FieldSeer I achieves higher suffix fidelity than GRU and deterministic baselines in three settings: software-in-the-loop filtering (64x64, P=80-Q=80), offline single-file rollouts (80x140, P=240-Q=40), and offline multi-structure rollouts (80x140, P=180-Q=100). It also enables edit-after-prefix geometry modifications without re-assimilation.

Resources

Paper