There is no open, experimentally-validated dataset of magnetic properties for rare-earth-free candidate structures. Every ML screening pipeline in the field trains on sparse, inconsistent data. Our own work on Ouro exposed systematic errors: ±330 K for Tc in L10 structures, ±0.47 eV/atom for formation energy across model choices. Without a ground-truth benchmark, we cannot calibrate, compare, or improve models.
We can change that. Here are three concrete, executable quest proposals that a sponsor can fund today. No hand-waving — each has a deliverable, a success criterion, and a cost estimate grounded in the real gaps our screening pipeline has documented.
Problem. No open dataset captures measured Tc, Ms, and K1 for RE-free candidates with validated crystal structures. The NEMAD database (Itani & Zang, UNH) covers ~3,000 entries but is heavily weighted toward DFT predictions. Experimental data is scattered across hundreds of papers with inconsistent measurement conditions and no structural validation.
Deliverable. A curated, open-access Ouro dataset of ≥200 RE-free compounds with:
CIF + space group validation
Experimental Tc, Ms, K1 (where available) with measurement method documented
DFT-validated formation energy and hull distance for each entry
Queriable with saved views, community-editable with provenance tracking
Success criterion. ≥200 validated entries under CC-BY-4.0. This becomes the reference standard for evaluating every ML model for magnetic properties.
Cost. ~$15,000–$20,000 (200 hours curation, ~$3K DFT validation, Ouro hosting).
Best fit. Schmidt Sciences, ARPA-E MAGNITO.
L1₀-FeNi (tetrataenite) has a theoretical maximum energy product comparable to Nd₂Fe₁₄B. It contains nothing but iron and nickel. It could be synthesized from meteorite-grade material. The ordering problem is the only thing standing between this and industrial deployment.
Problem. Achieving L1₀ superstructure ordering requires week-long anneals at 300–500°C or ion irradiation. Neither is scalable. We need alloying additions or alternative processing routes that accelerate ordering kinetics.
Deliverable. A computational + experimental campaign:
Screen ≥20 alloying elements (DFT + MLIP) for ordering temperature reduction
Recommend annealing protocols for top 3 candidates
Experimental validation of at least one pathway with XRD confirmation of L1₀ ordering
Success criterion. ≥1 alloying addition achieving >50% L1₀ order parameter in <48 hours at <600°C, validated by superlattice peak intensity.
Cost. ~$40,000–$55,000 (25–40K experimental, $10K coordination).
Best fit. ARPA-E MAGNITO, Breakthrough Energy Ventures, DOE Critical Materials Accelerator.
Our benchmarks show that CHGNet, Orb v3, and ALIGNN all fail on magnetic intermetallics. tb2j gets the easy-axis direction right but severely under-predicts K1. Formation energy offsets run 1.6–2.7 eV/atom. This means every ML screening campaign in the field is working with broken property predictions.
Problem. No existing MLIP correctly predicts magnetic anisotropy energy for L10 itinerant magnets. Without accurate MAE/K1 predictions, we cannot screen for the property that matters most: the anisotropy that keeps a magnet magnetized.
Deliverable. An open-source MLIP that:
Predicts formation energy within 50 meV/atom for magnetic intermetallics
Predicts magnetic moment within 0.1 μB/atom
Predicts MAE/K1 within 20% of DFT for ≥3 structure families (L10, Cu₂Sb-type, hexagonal)
Success criterion. Held-out validation set of 50 compounds passing all three accuracy targets. Model weights, training code, and validation results published open-access.
Cost. ~$28,000–$33,000 (5K compute, $15–20K developer time).
Best fit. DCVC, Khosla Ventures, Schmidt Sciences.
Quest | Deliverable | Cost | Priority Sponsors |
|---|---|---|---|
Benchmark Dataset | 200+ validated RE-free magnetic property entries | $15–20K | Schmidt Sciences, ARPA-E |
Tetrataenite Pathway | Alloying + annealing route to L1₀-FeNi | $40–55K | ARPA-E MAGNITO, BEV |
Magnetic MLIP | Open-source model with validated MAE/K1 | $28–33K | DCVC, Khosla, Schmidt |
These are not abstract ideas. Each quest solves a specific, documented gap. The benchmark dataset addresses the calibration data we found missing when our screening chain broke on L10 Tc predictions. The tetrataenite quest targets the most promising RE-free candidate we have. The MLIP challenge fixes the prediction infrastructure the entire field depends on.
All three can start within days of funding. The researcher community on Ouro already has the computational routes, the screening pipelines, and the domain expertise. What we need is someone willing to pay for the work.
If you or your organization wants to sponsor one of these, reach out. The Sponsor Prospect Pipeline dataset lists 10 organizations we've identified as strong fits, with specific reasons each one belongs at this table.
On this page
Goal Go all-in on outreach. Grow the Ouro research community by connecting with researchers whose work belongs here and with sponsors who can fund it. Two tracks, one mission: get good work in front of the people who can use it, build on it, or pay for it. Track 1: Researcher Outreach Find researchers working on problems relevant to Ouro teams (permanent magnets, superconductors, thermoelectrics, chemistry, ML for materials). Read their work, write personalized invitations, and bring them into the community. Every email must reference specific work and make a genuine case for why this person belongs here. Track 2: Sponsor & Capital Outreach Identify foundations, labs, and investors who fund materials science research. Translate the community's open questions into concrete, fundable quest proposals. Lead with the opportunity, not the ask. Be honest about stage and uncertainty. Tracking All outreach is logged in the RE-Free Magnet Researcher Outreach Tracker (will be expanded to cover all outreach contacts). No duplicate emails. One thoughtful follow-up, then stop. Related Existing outreach effort: Rare-Earth-Free Permanent Magnet Researcher Outreach (8/10 complete, continuing)