Institution profile

Proxima Fusion

Industry researcheurope · de
Official website
Research library1linked papers
Opportunities0open roles
Selected work

Representative Papers

ConStellaration: A dataset of QI-like stellarator plasma boundaries and optimization benchmarks

Jun 24, 2025

To address the lack of standardized optimization frameworks and high-quality benchmark datasets for quasi-isodynamic (QI) stellarator design, this work introduces the first publicly available, diverse dataset of QI stellarator boundary shapes, accompanied by ideal MHD equilibrium solutions and key performance metrics. We define three progressive optimization benchmark tasks—geometric constraint satisfaction, engineering feasibility, and multi-objective trade-off—and propose a unified generative framework integrating classical optimization, ideal MHD simulation, and data-driven modeling, enabling rapid generation of physically valid configurations without expensive physics simulations. This provides the first end-to-end open benchmark and strong baseline models for ML and optimization researchers, substantially lowering entry barriers. Experiments demonstrate that learned models efficiently generate novel, physically feasible stellarator configurations, accelerating interdisciplinary fusion energy research.

0 citationsRead paper
Recent publications

Latest Papers

ConStellaration: A dataset of QI-like stellarator plasma boundaries and optimization benchmarks

Jun 24, 2025

To address the lack of standardized optimization frameworks and high-quality benchmark datasets for quasi-isodynamic (QI) stellarator design, this work introduces the first publicly available, diverse dataset of QI stellarator boundary shapes, accompanied by ideal MHD equilibrium solutions and key performance metrics. We define three progressive optimization benchmark tasks—geometric constraint satisfaction, engineering feasibility, and multi-objective trade-off—and propose a unified generative framework integrating classical optimization, ideal MHD simulation, and data-driven modeling, enabling rapid generation of physically valid configurations without expensive physics simulations. This provides the first end-to-end open benchmark and strong baseline models for ML and optimization researchers, substantially lowering entry barriers. Experiments demonstrate that learned models efficiently generate novel, physically feasible stellarator configurations, accelerating interdisciplinary fusion energy research.

0 citationsRead paper