{
  "id": "2604.11791",
  "title": "A Mechanistic Analysis of Looped Reasoning Language Models",
  "authors": "Hugh Blayney, Álvaro Arroyo, Johan Obando-Ceron, Pablo Samuel Castro, Aaron Courville, Michael M. Bronstein, Xiaowen Dong",
  "first_public": "2026-04-13",
  "added_at": "2026-10-26",
  "venue": "arXiv",
  "arxiv": "2604.11791",
  "paper_url": "https://arxiv.org/abs/2604.11791",
  "html_url": "https://arxiv.org/html/2604.11791v1",
  "lens": [
    "recurrent",
    "analysis"
  ],
  "representation": [
    "hidden-state"
  ],
  "computation": [
    "recurrent",
    "depth-wise"
  ],
  "training": [
    "analysis-only",
    "pretraining"
  ],
  "questions": [
    "architecture",
    "interpretability",
    "generalization"
  ],
  "evidence": [
    "empirical",
    "mechanistic",
    "formal"
  ],
  "modality": [
    "language"
  ],
  "summary": "Analyzes how hidden states and attention change across repeated transformer layers. Finds that recurrent blocks often repeat the stages seen in feedforward models, while input injection and normalization affect whether those stages remain stable at greater depth.",
  "provenance": "Primary-source literature review, 2026-10-26",
  "author_names": [
    "Hugh Blayney",
    "Álvaro Arroyo",
    "Johan Obando-Ceron",
    "Pablo Samuel Castro",
    "Aaron Courville",
    "Michael M. Bronstein",
    "Xiaowen Dong"
  ],
  "record_url": "https://latent-implicit-thinking.github.io/papers/2604.11791/",
  "legacy_record_url": "https://latent-implicit-thinking.github.io/papers/#paper-2604-11791",
  "citation_date": "2026-04-13",
  "json_url": "https://latent-implicit-thinking.github.io/papers/2604.11791/record.json",
  "markdown_url": "https://latent-implicit-thinking.github.io/papers/2604.11791/index.md",
  "bibtex_url": "https://latent-implicit-thinking.github.io/papers/2604.11791/citation.bib",
  "csl_url": "https://latent-implicit-thinking.github.io/papers/2604.11791/citation.csl.json",
  "pdf_url": "https://arxiv.org/pdf/2604.11791"
}