{
  "abstract": "Adapting machine-learned interatomic potential (MLIP) foundation models to specialised tasks through fine-tuning is an increasingly important practice, yet systematic guidance on when and how to fine-tune is currently limited. We evaluate seven fine-tuning strategies -- naive full-parameter updates, two layer-freezing variants, Low-Rank Adaptation (LoRA), multihead replay, pseudolabelled replay, and replay combined with LoRA -- across five chemically diverse benchmarks (aqueous NaCl, ice polymorphs, S$_\\mathrm{N}$2 reactions, SPICE biomolecules, and lithium electrolytes), three generations of foundation models, and training sets spanning five orders of magnitude. To support this evaluation we implement three capabilities in the MACE codebase: LoRA adapted for equivariant message-passing architectures, including both scalar and equivariant linear layers; pseudolabelled replay, which decouples the replay data source from the original pretraining corpus; and model-aware atomic reference energy (E0) reestimation for fine-tuning workflows. We find that foundation model quality, correct E0 initialisation, and well-chosen hyperparameters are prerequisites whose impact routinely exceeds that of the fine-tuning strategy itself. Once these prerequisites are met, most strategies achieve strong target-task accuracy, consistently surpassing models trained from scratch. The practical distinction depends on deployment scope: naive fine-tuning offers the best convergence for single-system applications, while multihead replay -- with either original or pseudolabelled data -- is the only approach tested that consistently preserves out-of-distribution robustness, maintaining both pretraining-distribution accuracy for broader deployment and many-body short-range repulsion.",
  "abstract_shown": true,
  "affiliations": [
    "Department of Engineering , University of Cambridge , Cambridge , UK",
    "Scientific Computing Department , Science and Technology Facilities Council-Daresbury Laboratory , Warrington , UK",
    "Center for Materials Physics and Technology , U.S. Naval Research Laboratory , Washington , DC , USA"
  ],
  "article_type": null,
  "arxiv": null,
  "author_total": 6,
  "authors": [
    {
      "name": "Tamas Lajos Tompa"
    },
    {
      "affs": [
        0
      ],
      "name": "Eszter Varga-Umbrich",
      "orcid": "0000-0003-2871-8965"
    },
    {
      "affs": [
        0
      ],
      "name": "Ilyes Batatia",
      "orcid": "0000-0001-6915-9851"
    },
    {
      "affs": [
        1
      ],
      "name": "Alin M. Elena",
      "orcid": "0000-0002-7013-6670"
    },
    {
      "affs": [
        2
      ],
      "name": "Noam Bernstein",
      "orcid": "0000-0002-6532-1337"
    },
    {
      "affs": [
        0
      ],
      "name": "Gábor Cśanyi",
      "orcid": "0000-0002-8180-2034"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/3xvq2266vd0qa.bib",
  "cited_by": 21,
  "doi": "10.48550/arxiv.2606.12704",
  "fulltext_url": null,
  "id": "3xvq2266vd0qa",
  "issns": [
    "2331-8422"
  ],
  "landing_url": "https://arxiv.org/abs/2606.12704",
  "language": "en",
  "licence": "CC-BY",
  "licence_label": "CC BY",
  "licence_source": "journal",
  "licence_url": "https://creativecommons.org/licenses/by/4.0/",
  "metadata_licence": null,
  "native_subjects": [
    [
      "openalex",
      "2505|Machine Learning in Materials Science"
    ]
  ],
  "openalex": "W7164594465",
  "pdf_url": "https://arxiv.org/pdf/2606.12704",
  "pmcid": null,
  "pmid": null,
  "published": "2026-06-10",
  "publisher": "Cornell University",
  "ris": "https://research.draconitic.com/paper/3xvq2266vd0qa.ris",
  "sources": [
    {
      "first_seen": "2026-11-04T13:21:20Z",
      "native_id": "W7164594465",
      "source": "openalex",
      "url": "https://arxiv.org/abs/2606.12704"
    }
  ],
  "subjects": [
    {
      "field": "chemistry",
      "origin": "openalex:2505|Machine Learning in Materials Science",
      "subfield": "materials-polymers"
    }
  ],
  "title": "Fine-tuning MLIP foundation models: strategies for accuracy and transferability",
  "title_text": "Fine-tuning MLIP foundation models: strategies for accuracy and transferability",
  "type": "preprint",
  "updated_at": "2026-11-08T13:20:56Z",
  "url": "https://research.draconitic.com/paper/3xvq2266vd0qa",
  "venue": "arXiv (Cornell University)",
  "versions": []
}