{"id":"hyp-fe-persistent-outlier","title":"Fe Magnetic MLIP Failure Mode","subtitle":"The old PR > 2 trio claim is frozen after Born screening; Fe remains a spin/mechanical-stability target.","category":"conjectures","tags":["fe","outlier","mlip"],"source":"articles/docs/conjectures/fe-persistent-outlier.md","lang":"en","words":299,"readMinutes":1,"toc":[{"depth":2,"text":"Current status after Born screening","id":"current-status-after-born-screening"},{"depth":2,"text":"Claim","id":"claim"},{"depth":2,"text":"Historical evidence (pre-screening; do not cite as current PR evidence)","id":"historical-evidence-pre-screening-do-not-cite-as-current-pr-evidence"},{"depth":2,"text":"Why it stays open","id":"why-it-stays-open"},{"depth":2,"text":"Next","id":"next"}],"html":"<h1 id=\"fe-is-a-persistent-outlier\">Fe Is a Persistent Outlier</h1><p><strong>Status: Open — narrowed after Born-screened re-audit</strong></p>\n<h2 id=\"current-status-after-born-screening\">Current status after Born screening</h2><p>The old “Fe stays PR &gt; 2 under every foundation-MLIP addition” line is not\ncurrently citable. Born stability screening excludes <strong>CHGNet-Fe</strong> (C11 &lt; |C12|),\nso the original trio count used an invalid elastic tensor. Independently, at\nn = 8–11 Born-stable models per element, Fe&#39;s cross-model alignment is <strong>+0.80</strong>\nand the screened error axis remains one-dimensional rather than Fe-specific.</p>\n<p>What survives is narrower: Fe remains a high-priority magnetic failure-mode\nsuspect because one foundation model&#39;s Fe tensor fails mechanical stability and\nFe appears as a MACE-disagreement sentinel in later cross-MLIP analysis. That is\na mechanistic target, not the original PR-outlier proof.</p>\n<h2 id=\"claim\">Claim</h2><p>Iron (Fe) is a high-priority magnetic failure-mode suspect for foundation MLIPs.\nThe old claim that Fe is the one IMMI element that remains off-ribbon across the\nfoundation-model trio is frozen until Born-stable recomputation.</p>\n<h2 id=\"historical-evidence-pre-screening-do-not-cite-as-current-pr-evidence\">Historical evidence (pre-screening; do not cite as current PR evidence)</h2><p>Across the foundation-MLIP trio (MACE-MP-0, CHGNet, Orb-v3), Fe was the held-out\n1 of 15 in every pre-screening &quot;14/15 on the ribbon&quot; result. Its participation\nratio appeared to stay <strong>PR &gt; 2</strong> regardless of LAM addition. This paragraph is\npreserved for provenance, but the current claim must be recomputed after Born\nscreening.</p>\n<h2 id=\"why-it-stays-open\">Why it stays open</h2><p>The original invariance claim is not well-established after screening; the\n<em>failure mode</em> remains compelling. Fe&#39;s magnetism (BCC ferromagnetic ground\nstate, complex spin-dependent bonding) is the obvious suspect, but that is a\nhypothesis, not a result. It is also entangled with the <a href=\"#/read/hyp-bccfcc-causal-shield\">BCC &gt; FCC residual</a>\nand the <a href=\"#/read/hyp-au-mlip-escape\">Au escape</a> — the three &quot;special element&quot; findings may\nshare a mechanism.</p>\n<h2 id=\"next\">Next</h2><p>Test the magnetism hypothesis explicitly: stratify Fe potentials by whether they\nmodel spin, run Born stability screening before PR calculations, and check\nwhether spin-aware references remove the mechanical-stability failure mode.</p>\n"}