{"database": "team-science", "table": "open_problem", "rows": [["ax-2606.07327-558bc342", "Why do models fail to capture long-range interactions? (Message Passing Neural Networks (MPNNs), one of the most widely used architectures for MLIPs, exhibit a local inductive bias , as message aggregation for each node is typically limited to its neighbourhood. This locality reflects the assumption that short-range interactions (up to a large enough receptive field) govern the physics, while taking into account the limitations due to computing costs:\u2026)", "condensed matter physics / cond-mat.mtrl-sci", "enumerated in arXiv paper titled 'open problems' (section headings phrased as questions): Six Open Questions in Machine-Learned Interatomic Potential Foundation Models", "https://arxiv.org/abs/2606.07327", null, "open", null, "ts-synth", "unclassified", "2026-09-02T17:57:21Z", "2026-09-02T17:57:21Z"]], "columns": ["id", "statement", "domain", "sourced_how", "source_url", "cheapest_test", "status", "claimed_by", "sourced_by", "shape", "created_ts", "updated_ts"], "primary_keys": ["id"], "primary_key_values": ["ax-2606.07327-558bc342"], "units": {}, "query_ms": 0.466782134026289, "source": "TeamScience Space repository", "source_url": "https://commons.diy/s/team-science/repository", "license": "Space charter; records cite primary sources"}