electromagnetic simulation

To address the challenge of efficiently generating large-scale, high-fidelity electromagnetic (EM) simulation data required for machine learning in antenna design, this work develops an open-source, GPU-accelerated EM simulation framework built upon gprMax. By deeply integrating CUDA parallel computing, the framework achieves substantial throughput improvement while maintaining sub-millimeter spatial resolution; its accuracy is experimentally validated to match that of leading commercial EM solvers. The framework further incorporates diverse machine learning and deep learning models to enable high-accuracy inverse prediction—from EM responses to antenna geometry and material parameters. Experimental results demonstrate speedups of over 5× on entry-level GPUs versus high-end CPUs, and up to 18× on flagship gaming GPUs, drastically reducing dataset generation time. This work establishes a reproducible, scalable, and cost-effective simulation–learning infrastructure for data-driven intelligent antenna design.

This work addresses the long-horizon electromagnetic field dynamics prediction problem for 2D TE waveguides under partial observability. We propose a geometry-aware world model that integrates structural/material map guidance, source-action conditional generation, and a symmetric logarithmic-domain GRU architecture, enabling closed-loop rollout and physics-domain interactive simulation on FDTD-based benchmarks. Our key contribution is the first demonstration of real-time geometric editing—after prefix observation—without reinitializing data assimilation, substantially enhancing interactivity and design adaptability in digital twin systems. Evaluated across three canonical photonic scenarios, the model consistently outperforms GRU and deterministic baselines across 200 reproducible simulations, achieving markedly improved fidelity in suffix-field prediction. This establishes a new, efficient, and editable dynamic modeling paradigm for inverse design of photonic devices.

To address the challenge of real-time, high-fidelity electromagnetic (EM) propagation simulation in complex 3D indoor environments, this paper proposes a theory-driven conditional generative adversarial network (cGAN). For the first time, physics-based EM priors—including path loss models and specular reflection constraints—are explicitly embedded into the cGAN architecture; geometric scene structure and transmitter location are jointly encoded, and the model is supervised using ray-tracing-generated ground-truth data. This work presents the first end-to-end framework capable of generating real-time EM field-strength heatmaps for intricate 3D indoor scenes. Evaluated across 19 real-world environments, it achieves accuracy comparable to ray tracing—demonstrating lower mean squared error (MSE)—while accelerating inference by 5×, with single-prediction latency in the millisecond range. The core contribution lies in unifying physical interpretability with deep learning efficiency, thereby significantly expediting the deployment and optimization of wireless communication systems.

This study addresses the high computational cost and steep learning curve associated with electromagnetic simulation modeling. We propose a conversational, LLM-driven automation framework for electromagnetic simulation, centered on Google Gemini 2.0 Flash as the reasoning engine and tightly integrated with Gmsh (geometry generation), GetDP (finite-element solver), and Python (workflow orchestration). The method enables users to specify physical configurations, boundary conditions, and solution objectives via natural language, automatically generating executable simulation scripts, solving 2D eddy-current problems, and delivering customized post-processing and concise result summaries. Key contributions include the first end-to-end, natural-language-to-simulation pipeline for electromagnetic analysis and seamless integration of domain-specific solvers with modern LLMs. Experimental evaluation across diverse conductor geometries demonstrates high modeling accuracy, strong generalization capability, and substantial reductions in user expertise requirements—thereby improving modeling efficiency and human–machine interaction.

Wireless channel modeling for communications and radar systems suffers from heavy reliance on high-quality labeled data, poor generalization, and limited physical interpretability. To address these challenges, this paper proposes a Sparse Bayesian Generative Modeling (SBGM) framework that explicitly incorporates physical priors. Specifically, it is the first to embed the inherent compressibility of wireless channels into a generative model, enabling online learning from compressed measurements. Physical constraints—derived from electromagnetic propagation principles—and parameterized channel representations are integrated to ensure model transparency and interpretability. Moreover, the method supports zero-shot transfer across antenna configurations and frequency bands without retraining. Experimental results demonstrate that SBGM achieves high-fidelity reconstruction of channel parameter distributions using only a small number of compressed measurements. This significantly reduces data acquisition and labeling overhead while markedly improving environmental adaptability and cross-scenario generalization performance.

This work addresses the frequent mismatch between user-specified physical intent and the actual behavior of multiphysics simulation code generated by large language models, often due to erroneous implementations of partial differential equations (PDEs). To bridge this gap, we propose a PDE-structure-based intent verification method that deterministically reconstructs the governing equations implicitly encoded in the generated code and compares them against the user’s intended PDEs, enabling semantic correctness validation and iterative refinement. We introduce, for the first time, a formal metric termed the Intent Fidelity Score (IFS) to quantify alignment with physical intent, establish a PDE-driven feedback loop, and demonstrate compatibility with major PDE frameworks including MOOSE, FEniCS, and FreeFEM. Evaluated on 220 cases in MooseBench, our approach substantially improves IFS—by 0.22–0.41 on challenging instances with initial IFS < 0.7—while audits reveal that execution-only repair strategies still yield physically incorrect results in 39–40% of cases.

This work addresses the limited accuracy of conventional radiomap modeling at high frequencies, which stems from neglecting fine-scale scatterers. The authors propose a multi-scatterer channel model based on spherical wave modal expansion that unifies the characterization of source radiation, single scattering, and multiple-scattering coupling effects through modal superposition. By reformulating the forward model as an inverse optimization problem, the approach jointly estimates scatterer responses and transmitter location. Notably, it integrates multi-scattering interactions with low-order modal approximation within a physically interpretable framework—a first in the field—and enables high-fidelity radiomap reconstruction and extrapolation from sparse measurements. Simulations demonstrate that the proposed model significantly outperforms existing methods in both spatial and beam domains, particularly in dense scattering environments.

This work proposes a scalable parallel interpolation algorithm to address the challenge of data consistency across multiple solvers in overlapping grid regions for atmospheric wave simulations. The method enables efficient, in situ data coupling between solvers by introducing a grid-coupling mechanism tailored for large-scale parallel environments, ensuring global solution consistency while balancing computational efficiency and communication overhead. Evaluated in realistic atmospheric wave simulation scenarios, the algorithm demonstrates strong scalability and high performance, significantly improving both the accuracy and efficiency of data exchange among multiphysics solvers.

This work addresses the challenges of integrating evolutionary computation into scientific instrument design, which is hindered by highly constrained design spaces, vast search dimensions, and expensive physical simulations. To overcome these obstacles, the authors propose a modular evolutionary computation framework tailored for scientific engineering applications. The framework innovatively combines domain-aware geometric-parametric individual encoding, seamless integration with external simulators, and an evolution strategy specifically adapted to low-throughput evaluation environments. Its effectiveness and practicality are demonstrated through successful applications in space science, including the optimization of 3D antennas and aerodynamic shapes for very low Earth orbit spacecraft, where it significantly reduces drag in real-world, high-fidelity simulation scenarios.

This work addresses the lack of lightweight, customizable, and easily deployable web-based solutions for graphical EMF model editing, as existing tools predominantly rely on desktop environments. The paper proposes EMFular, a novel front-end-only framework that automatically generates semantic-compliant graphical editors directly from Ecore metamodels without requiring a back-end. By integrating Angular’s front-end architecture, EMF semantics-preserving mechanisms, and standardized serialization techniques, EMFular supports full modeling operations and seamless interoperability with the EMF toolchain. High customizability is achieved through Angular extension points. Experimental results demonstrate that the generated editors exhibit strong functional completeness, adaptability, and robustness, significantly lowering the barriers to deployment and customization of EMF-based modeling in web environments.