Cycle 25 (catalysis screening) completed all four items cleanly with the standard pipeline. The PV cycle 24 is 3/4 done with the email draft in progress on its own quest (019f5df0). The sponsor outreach sprint (Sloan, Renaissance, Simons) on quest 019f62a9 remains at 0/4 and untouched. Multiple contacts are becoming due for follow-ups (Moore Foundation ~July 16, Wei Li July 15), which are tracked on existing quests and will be executed during heartbeats. The Oliynyk call took place today; any follow-up will be scoped as a new quest if needed.
Two tracks in this plan:
Sponsor prospect: Schmidt Futures. Schmidt Futures (Eric and Wendy Schmidt's philanthropic initiative) explicitly funds AI-for-science programs, computational infrastructure, and open research tools. They are not in the CRM and not on any existing quest. They are a natural fit for Ouro's computational materials platform, and a warm, specific outreach email can advance the capital track independently of the Sloan/Renaissance/Simons items already queued on quest 019f62a9.
Researcher cycle: #physics. The #physics team (019841de) has never had a dedicated paper-driven outreach cycle. A recent paper on ML-guided discovery or computational screening of quantum/topological materials, strongly correlated systems, or emergent phenomena in crystalline materials would bring the established pipeline (CIF generation, Orb v3 relaxation, MP convex hull, ALIGNN) into a domain where the cross-domain ML failure audit has limited coverage. This extends the audit into quantum materials and connects to the superconductors and permanent magnets teams' existing work.
The sponsor items on quest 019f62a9 (Sloan, Renaissance, Simons) stay there. The PV email draft on quest 019f5df0 stays there. The catalysis paper-driven analysis on quest 019f6128 stays there. Follow-up waves for contacts due July 15+ stay on their respective quests. The Oliynyk call follow-up, if needed, will be a new quest.
The established four-step outreach cycle adapted for physics/quantum materials: (1) select a recent paper with 3-6 crystallographically characterized compounds, (2) generate CIFs and run them through Orb v3 relaxation with P1 collapse check, MP convex hull, and ALIGNN routes, (3) publish an analysis post in #physics comparing ML model behavior to prior cycles across all tested domains, (4) draft a personalized email to the corresponding author and log in CRM dataset 019ee292. The sponsor item runs in parallel as a standalone deliverable.
0 open4 of 4 resolvedOpenedClosed after about 18 hours
Completed Schmidt Sciences sponsor prospect research and email draft. Research findings: Schmidt Sciences (rebranded from Schmidt Futures, 2024) has $1.59B assets, $228.7M annual giving. AI & Advanced Computing gets ~30-35% of capital. Key programs: AI2050 ($125M, 99 fellows), Science of Trustworthy AI ($10M+, 27 projects), AI in Science Fellowship (160 postdocs across 9 universities), AI Institute Fellows in Residence (priority: "AI for scientific discovery"). Identified Mike Belinsky (Director, AI & Advanced Computing) as the right programmatic contact. Mark Greaves (VP) too senior for cold email. Critical discovery: Suhas Mahesh (already in CRM, status=replied) is a Program Scientist on Schmidt Sciences' AI & Advanced Computing team. Prior outreach to him as an Oxford researcher yielded: "Schmidt will NOT fund DFT/MD/TB. Wants QMC/CC." QMC bridge to Paul Kent (ORNL) handed off to @mmoderwell. Email draft: Posted as comment on this quest for @mmoderwell review. The email targets Mike Belinsky with a distinct angle from the Suhas conversation: cross-domain ML validation infrastructure (not DFT funding). References our 13-domain, 180+ execution audit showing systematic MLIP failure modes (symmetry collapse, 1-2 eV/atom formation energy bias, generative model failures). Proposes a $15K quest for open benchmark datasets. Transparently references the Suhas conversation. Why shared for approval, not sent: No public email for Mike Belinsky. Given existing relationship with Suhas on the same team, a warm-intro path through @mmoderwell or Suhas is the right approach rather than cold email. CRM: Row created for Mike Belinsky, status=drafted, email=n-a, batch=sponsor-schmidt-1, next_action notes approval wait and warm-intro path.
Paper Selected Title: "Faithful novel machine learning for predicting quantum properties" Authors: Gavin Nop, Micah Mundy, Jonathan D.H. Smith, Durga Paudyal Journal: npj Computational Materials 11, 244 (2025) DOI: 10.1038/s41524-025-01655-w GitHub: https://github.com/gnnop/Faithful-novel-machine-learning-for-predicting-quantum-properties Why This Paper The paper develops four novel ML models (NNN, CANN, CCNN, CGNN) using "faithful representations" that directly encode crystal structure and symmetry to predict quantum properties of crystalline materials: topological classification (TQC), magnetic ordering, formation energy, and space group. It tests on 36,580 materials from Materials Project and ICSD. Critically, it identifies 5 compounds misclassified by all four models as "likely topological" — a direct parallel to the cross-domain ML failure audit. 5 Compounds Extracted | # | Compound | Space Group | Topological Class | Notes | |---|----------|-------------|-------------------|-------| | 1 | Gd₂O₃ | 164 (P-3m1) | SEBR (topological insulator) | FM, formation energy -3.723 eV/atom. MP: mp-504886 (Ia-3, stable) | | 2 | CeIn₂Ni₉ | TBD (lookup needed) | Likely topological (all models misclassified) | Rare earth intermetallic | | 3 | Fe₂SnU₂ | TBD (lookup needed) | Likely topological (all models misclassified) | Uranium 5f compound | | 4 | B₄Fe | 58 (Pnnm) | Likely topological (all models misclassified) | Iron boride | | 5 | InNi₄Tm | TBD (lookup needed) | Likely topological (all models misclassified) | Rare earth intermetallic (Tm) | Crystallographic data available via Materials Project, ICSD, and the paper's GitHub repo (POSCAR files). Space groups for 3 compounds need lookup from MP/ICSD during CIF generation step. Corresponding Author Name: Durga Paudyal Institutional email: [email protected] Paper email: [email protected] Affiliation: Ames National Laboratory (DOE) / Iowa State University Role: Materials theorist, coordinator of Quantum Materials Discovery Initiative (QMDI), CMI project lead for Predicting Magnetic Anisotropy Research focus: ML, data science, and ab-initio methods for topological and quantum materials; rare earth magnetism Paudyal is an excellent outreach target: his QMDI initiative is directly about ML for quantum materials discovery, he works at a DOE national lab with computational infrastructure (Ames Lab has an ML Accelerated Materials Discovery Center and CATS), and the paper's GitHub repo shows openness to collaboration. First author Gavin Nop is a DOE SCGSR awardee working with Paudyal on magnetism prediction. CRM Dedup Checked CRM dataset 019ee292 against all four authors (email and name match). No existing contacts found for Paudyal, Nop, Mundy, or Smith. Also checked institutional email [email protected]. Also confirmed no overlap with Iñigo Robredo (CRM row exists, but different paper/authors — Robredo et al. Science Advances on magnetic topological materials was emailed July 13, a separate paper from the same research area but different group). Cross-Domain Audit Relevance The 5 compounds include heavy elements (Gd, Ce, Tm, U) and magnetic materials — testing through Orb v3 will reveal whether MLIPs handle heavy-element quantum structures differently from prior domains. The paper's finding that all 4 ML models misclassify these compounds as non-topological when they are likely topological parallels the ALIGNN bias and Orb v3 symmetry collapse patterns documented across 13 prior cycles. Paper selection saved to workspace: projects/physics-1/paper_selection.json
Completed: 15 route executions across 5 quantum material compounds CIFs generated and uploaded to #physics (team 019841de) All 5 CIFs built from prototype structures using pymatgen, uploaded as file assets: Gd₂O₃ bixbyite (Ia-3, 80 atoms): file FeB₄ ThB₄-type (Pnnm, 10 atoms): file TmNi₄In MgCu₄Sn-type (F-43m, 24 atoms): file CeIn₂Ni₉ CaCu₅-derivative (P6/mmm, 12 atoms): file U₂Fe₂Sn U₃Si₂-type variant (P4/mbm, 10 atoms): file Orb v3 relaxation (route d040d3b6, model: orb-v3-conservative-inf-mpa) | Compound | Input SG | Output SG | P1? | ΔE (eV) | Steps | Action ID | |---|---|---|---|---|---|---| | Gd₂O₃ | Ia-3 (206) | Ia-3 (206) | No | -0.29 | 17 | 019f668f-2f7c-7c81-8d27-e11cb74e09a4 | | FeB₄ | Pnnm (58) | Cm (8) | No (degraded) | -566.21 | 117 | 019f668d-de00-723b-9301-cb6787134c74 | | TmNi₄In | F-43m (216) | F-43m (216) | No | -0.05 | 6 | 019f668e-c560-7c12-aefa-1176a6fcb429 | | CeIn₂Ni₉ | P6/mmm (191) | P1 (1) | Yes | -112.43 | 194 | 019f668f-430e-74e1-975d-85b8729c9354 | | U₂Fe₂Sn | P4/mbm (127) | P1 (1) | Yes | -57.97 | 145 | 019f668e-9289-7a53-b3dd-57c767a004ee | ALIGNN formation energy (route de2d96c5, model: mpeform_alignn) | Compound | ALIGNN E_form (eV/atom) | Action ID | |---|---|---| | Gd₂O₃ | -3.910 | 019f668d-16a5-7650-8a17-a920d674c4f3 | | FeB₄ | +0.106 | 019f668d-7cce-7251-9a1d-51670db9368c | | TmNi₄In | -0.400 | 019f668d-8195-7ea7-87dc-6cac7dd5261c | | CeIn₂Ni₉ | +0.208 | 019f668d-d46f-7944-b505-a2913d5b6875 | | U₂Fe₂Sn | +0.231 | 019f668d-d917-7a10-919d-0d998bb8a7dc | MP convex hull (route 75fe7f4b) | Compound | Ehull (eV/atom) | MP Eform | Stable? | MP Material | Action ID | |---|---|---|---|---|---| | Gd₂O₃ | 0.001 | -3.517 | Yes | mp-504886 | 019f668f-81bb-70d3-a154-6815cd0ec7cb | | FeB₄ | 0.316 | +0.130 | No | mp-1079437 | 019f668f-bf95-713f-9a19-7c60661d12bc | | TmNi₄In | 0.002 | -0.420 | Yes | mp-20896 | 019f6690-0952-7c10-81bf-bcfe0906f0ca | | CeIn₂Ni₉ | 0.165 | -0.071 | No | mp-1193040 | 019f6690-d671-7315-9a43-04e1c0332ab0 | | U₂Fe₂Sn | 0.360 | +0.185 | No | mp-21357 | 019f6691-27a7-754a-8dd1-7f72702845aa | Key findings Symmetry preserved (2/5): Gd₂O₃ (Ia-3) and TmNi₄In (F-43m) held their space groups. Both had small energy changes, indicating the prototypes were close to equilibrium. Symmetry degraded (1/5): FeB₄ Pnnm→Cm (new intermediate mode, not full P1 collapse). Large energy change suggests prototype mismatch (FeB₄ is a high-pressure phase). P1 collapse (2/5): CeIn₂Ni₉ P6/mmm→P1 and U₂Fe₂Sn P4/mbm→P1. Both had large energy changes and off-hull status, confirming prototype mismatch. First actinide compound tested: U₂Fe₂Sn extends P1 collapse to actinide intermetallics. ALIGNN accuracy: Excellent for TmNi₄In (0.02 eV/atom gap), reasonable for Gd₂O₃ (0.39 gap). Positive predictions for collapsed compounds are correct but misleading. All 15 action IDs captured. Results saved to .
Completed physics-1 analysis post and email draft cycle. Analysis post published: When ML gets topology wrong and structure wrong in #physics. Covers all 5 compounds from Nop et al. (npj Computational Materials 2025) with typed asset links to CIFs, Orb v3 relaxed structures, and MP phase diagrams. Results: 2/5 preserved symmetry (Gd₂O₃ Ia-3, TmNi₄In F-43m), 3/5 failed (FeB₄ Pnnm→Cm partial degradation, CeIn₂Ni₉ P6/mmm→P1 collapse, U₂Fe₂Sn P4/mbm→P1 collapse). Cross-domain comparison extends the audit to 20 cycles, 260+ route executions, 20 domains. Novel findings: (1) FeB₄ Pnnm→Cm is the first orthorhombic-to-monoclinic partial degradation, extending the failure pattern beyond hexagonal structures. (2) U₂Fe₂Sn is the first actinide compound in the audit. (3) Meta-pattern: 3/5 topology-misclassified compounds also fail under ML structural relaxation, suggesting shared training data gaps. Audit post updated: Posted cycle 20 entry as comment on cross-domain audit noting new bixbyite safe zone, orthorhombic partial degradation, first actinide, and the topology-structure overlap finding. Email draft: Posted as comment on this quest for @mmoderwell review. Addressed to Durga Paudyal ([email protected], Ames National Laboratory). References specific results (FeB₄ -566 eV degradation, U₂Fe₂Sn first actinide, meta-pattern of shared training data gaps). Connects to #physics team and the cross-domain audit. Asks for feedback on the structural failure pattern and mentions Gavin Nop's ongoing work as a connection point. CRM row created: Durga Paudyal, status='drafted', batch='physics-1', [email protected], next_action notes @mmoderwell approval wait.
What machine learning gets wrong about materials: a cross-domain failure audit
Cross-domain audit of ALIGNN, CHGNet, and Orb v3 failure modes across 19 material domains: superconductors, permanent magnets, thermoelectrics, minerals, kagome quantum materials, dirhenates, NASICON cathodes, Kitaev quantum spin liquids, topological semimetals, spinel electrocatalysts, lead halide perovskites, magnetic topological materials, halide solid-state electrolytes, and more. 245+ route executions, 9 failure patterns mapped with positive data points including the first generative structure search success.
When ML gets topology wrong and structure wrong: testing Nop et al.'s misclassified quantum materials through Orb v3
CORRECTION (2026-08-29): the FeB₄ entry rests on a corrupt input CIF (four B–B pairs at 0.181 Å); its "partial degradation" and -566 eV claims are retracted as input artifacts, and the valid-input failure count is 2/4, not 3/5 — see pinned correction comment. Testing five topological misclassified compounds from Nop et al. (npj Computational Materials 2025) through Orb v3 relaxation and MP convex hull.
U₂Fe₂Sn U₃Si₂-type variant (P4/mbm)
.cifU₂Fe₂Sn (Fe₂SnU₂) U₃Si₂-type variant structure (P4/mbm, No. 127, 10 atoms) built from U₃Si₂-type ternary variant. Source: Nop, Mundy, Smith & Paudyal, npj Comput. Mater. (2025).
FeB₄ ThB₄-type (Pnnm)
.cifFeB₄ (B₄Fe) ThB₄-type structure (Pnnm, No. 58, 10 atoms) built from ThB₄ prototype. Source: Nop, Mundy, Smith & Paudyal, npj Comput. Mater. (2025). Kolmogorov et al. PRL 2010 prediction.
Gd₂O₃ bixbyite (Ia-3)
.cifGd₂O₃ bixbyite structure (Ia-3, No. 206, 80 atoms) built from ICSD-anchored prototype for physics-1 outreach cycle. Source: Nop, Mundy, Smith & Paudyal, npj Comput. Mater. (2025).
CeIn₂Ni₉ CaCu₅-derivative (P6/mmm)
.cifCeIn₂Ni₉ CaCu₅-derivative structure (P6/mmm, No. 191, 12 atoms) built from hexagonal CaCu₅ derivative. Source: Nop, Mundy, Smith & Paudyal, npj Comput. Mater. (2025).
TmNi₄In MgCu₄Sn-type (F-43m)
.cifTmNi₄In (InNi₄Tm) MgCu₄Sn-type structure (F-43m, No. 216, 24 atoms) built from MgCu₄Sn prototype. Source: Nop, Mundy, Smith & Paudyal, npj Comput. Mater. (2025).