{
  "abstract": "Learned temporal models are moving beyond offline forecasting toward interactive environments for decision making, yet logged-trajectory accuracy need not predict closed-loop behavior under policies optimized in the learned model. This forecasting-control gap matters in tokamak magnetic control, where coupled plasma-circuit dynamics and actuator constraints can amplify errors through feedback. Using a calibrated free-boundary Grad-Shafranov solver, we collect four datasets in a 2x2 design over physics-parameter and current-reference coverage. At matched budgets of 500, 1k, 3k, and 5k rollouts, we train 16 action-conditioned Transformers; PPO policies optimized in the 5k surrogates are then evaluated without adaptation on six reference-solver conditions. Only joint-coverage D4 shows monotonically decreasing cross-condition mean error, whereas D1 improves in distribution but degrades out of distribution. Binary completion masks substantial closed-loop differences: D2-D4 each complete 83.3% of conditions, but reach on average 99.0%, 94.7%, and 90.5% of the evaluation horizon, while their worst conditions reach 94.2%, 68.2%, and 43.2%. Policies trained with action noise complete 5/24 matched conditions, compared with 19/24 under nominal training, and have a lower mean horizon-capped episode length. Based on one trained policy per configuration and one evaluation realization per condition, the results support combining prediction error, episode completion, condition-level duration, and tracking quality.",
  "abstract_shown": true,
  "affiliations": [],
  "article_type": null,
  "arxiv": "2611.03690",
  "author_total": 10,
  "authors": [
    {
      "name": "Guoyang Shi"
    },
    {
      "name": "Siqi Ding"
    },
    {
      "name": "Xuanhe Wang"
    },
    {
      "name": "Pei Guo"
    },
    {
      "name": "Changquan Yu"
    },
    {
      "name": "Jiale Hu"
    },
    {
      "name": "Jiayi Zhi"
    },
    {
      "name": "Zhengyuan Chen"
    },
    {
      "name": "Jianguo Chen"
    },
    {
      "name": "Tianyuan Liu"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/b4q6vdvf9ys2y.bib",
  "cited_by": 0,
  "doi": "10.48550/arxiv.2611.03690",
  "fulltext_url": null,
  "id": "b4q6vdvf9ys2y",
  "issns": [],
  "landing_url": "https://arxiv.org/abs/2611.03690",
  "language": null,
  "licence": "CC-BY-4.0",
  "licence_label": "CC BY 4.0",
  "licence_source": "article",
  "licence_url": "http://creativecommons.org/licenses/by/4.0/",
  "metadata_licence": "CC0-1.0",
  "native_subjects": [
    [
      "arxiv",
      "cs.AI"
    ],
    [
      "arxiv",
      "cs.LG"
    ],
    [
      "arxiv",
      "physics.plasm-ph"
    ]
  ],
  "openalex": null,
  "pdf_url": "https://arxiv.org/pdf/2611.03690",
  "pmcid": null,
  "pmid": null,
  "published": "2026-11-01",
  "publisher": null,
  "ris": "https://research.draconitic.com/paper/b4q6vdvf9ys2y.ris",
  "sources": [
    {
      "first_seen": "2026-11-04T15:21:25Z",
      "native_id": "2611.03690",
      "source": "arxiv",
      "url": "https://arxiv.org/abs/2611.03690"
    }
  ],
  "subjects": [
    {
      "field": "physics",
      "origin": "arxiv:physics.plasm-ph",
      "subfield": "plasma"
    }
  ],
  "title": "From Forecasting to Feedback: Evaluating Learned Dynamics for Tokamak Control",
  "title_text": "From Forecasting to Feedback: Evaluating Learned Dynamics for Tokamak Control",
  "type": "preprint",
  "updated_at": "2026-11-06T18:21:01Z",
  "url": "https://research.draconitic.com/paper/b4q6vdvf9ys2y",
  "venue": "arXiv",
  "versions": []
}