{
  "abstract": "We separate the extrapolation problem in many-body systems into two axes: toward combinations with more components, and toward components unseen. The prevailing pipeline—encode then interact—performs poorly when extrapolating in composition count. We propose a route that expresses interaction without an interaction module: a per-component encoder shaped by the task compresses interaction behaviour into each component’s encoding, and the encodings are combined and read out by a simple MLP. Where component identities are bounded while their combinations are not, this route performs better along the composition axis—a difference that follows from the underlying function form rather than from the data. Chemical mixtures are the main example, synthetic tasks isolate the two axes, and the same structure recurs in other fields.",
  "abstract_shown": true,
  "affiliations": [
    "Zhengzhou, China, 450001"
  ],
  "article_type": null,
  "arxiv": null,
  "author_total": 10,
  "authors": [
    {
      "affs": [
        0
      ],
      "name": "Xiao Yang",
      "orcid": "0000-0002-0159-1434"
    },
    {
      "affs": [
        0
      ],
      "name": "Guanghao Song"
    },
    {
      "affs": [
        0
      ],
      "name": "Feiyu Li"
    },
    {
      "affs": [
        0
      ],
      "name": "Zihan Xu"
    },
    {
      "affs": [
        0
      ],
      "name": "Yanan Yang"
    },
    {
      "affs": [
        0
      ],
      "name": "Baojiang He"
    },
    {
      "affs": [
        0
      ],
      "name": "Lingbo Ji"
    },
    {
      "affs": [
        0
      ],
      "name": "Xiankuan Huo"
    },
    {
      "affs": [
        0
      ],
      "name": "Jun Hu"
    },
    {
      "affs": [
        0
      ],
      "name": "Zhan Qu"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/6z6bhydaa642w.bib",
  "cited_by": 0,
  "doi": "10.26434/chemrxiv.15010330/v1",
  "fulltext_url": null,
  "id": "6z6bhydaa642w",
  "issns": [],
  "landing_url": "https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010330/v1",
  "language": null,
  "licence": "CC-BY-NC",
  "licence_label": "CC BY-NC",
  "licence_source": "journal",
  "licence_url": "https://creativecommons.org/licenses/by-nc/4.0/",
  "metadata_licence": null,
  "native_subjects": [
    [
      "openalex",
      "2505|Machine Learning in Materials Science"
    ]
  ],
  "openalex": "W7221588494",
  "pdf_url": "https://chemrxiv.org/doi/pdf/10.26434/chemrxiv.15010330/v1",
  "pmcid": null,
  "pmid": null,
  "published": "2026-10-13",
  "publisher": "American Chemical Society (ACS)",
  "ris": "https://research.draconitic.com/paper/6z6bhydaa642w.ris",
  "sources": [
    {
      "first_seen": "2026-10-30T12:17:24Z",
      "native_id": "10.26434/chemrxiv.15010330",
      "source": "chemrxiv",
      "url": "https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010330/v7"
    }
  ],
  "subjects": [
    {
      "field": "chemistry",
      "origin": "openalex:2505|Machine Learning in Materials Science",
      "subfield": "materials-polymers"
    }
  ],
  "title": "Two extrapolation axes for many-body systems: interaction without an interaction module, chemical mixtures as the main example",
  "title_text": "Two extrapolation axes for many-body systems: interaction without an interaction module, chemical mixtures as the main example",
  "type": "preprint",
  "updated_at": "2026-11-06T12:20:43Z",
  "url": "https://research.draconitic.com/paper/6z6bhydaa642w",
  "venue": "ChemRxiv",
  "versions": []
}