Ouro has exactly one MAE route: Calculate magnetic anisotropy energy, a DFT calculation using the TB2J method. It's a real first-principles calculation — spin-orbit coupling, non-collinear DFT, direction-dependent total energies. The results are trustworthy. But it costs $4.00 per call and takes minutes to hours per structure, depending on convergence parameters and system size.
For systematic screening, that's a non-starter. A typical permanent-magnet discovery campaign needs to evaluate hundreds to thousands of candidate compositions. At $4/call with DFT wall-times, that's thousands of dollars and weeks of compute. We've been running our screening pipeline with formation energy (ALIGNN), structure relaxation (Orb v3), convex hull (MP), and magnetic moment (CHGNet) — all fast ML or pre-computed routes. MAE is the one property we can't screen at scale, and it's arguably the most important: a material with high magnetization but low anisotropy won't hold its magnetization at room temperature. K1 is the figure of merit that separates a permanent magnet from a soft magnet.
Recent literature shows ML can fill this gap. Navrátil et al. (npj Computational Materials, 2025) demonstrated interpretable ML for atomic-scale magnetic anisotropy prediction from crystal structure alone. A CGCNN-based approach was used to predict anisotropy constants in Fe/Ni-substituted MnAl systems (arXiv:2404.03051). The Fe-Co-X search (J. Magn. Magn. Mater., 2026) integrated ML screening with DFT-validated MAE. The pattern is consistent: train on DFT-computed MAE datasets, predict at GNN speed.
Primary input: CIF file (Ouro file asset, same as the existing DFT route).
Additional parameters:
spin_axis (optional, default auto): Crystallographic direction for initial spin alignment, e.g. [001], [100], [110]. If auto, the model predicts the easy axis from structure alone.
magnetic_config (optional, enum: fm, afm, ferrimagnetic, auto): Initial magnetic ordering hint. If auto, the model infers likely ordering from composition and structure.
temperature (optional, float, default 0 K): For models that include thermal effects. Most current ML approaches are 0 K; this is a forward-looking parameter.
The key design choice: the route should accept a bare CIF and return a prediction without requiring the user to specify spin direction or magnetic configuration. That's what makes it a screening tool rather than a calculation tool. The existing DFT route already requires the CIF as input_assets.file, so the interface pattern is established.
{ "compound": "MnAl", "space_group": "P4/mmm", "mae_mev_per_atom": 0.275, "mae_mj_per_m3": 1.18, "k1_constant": 1.18, "easy_axis": [0, 0, 1], "hard_axis": [1, 0, 0], "confidence": 0.85, "method": "ML_surrogate", "dft_reference_mae": null, "notes": "Predicted from crystal structure. Recommend DFT validation for top candidates." }
The critical fields:
MAE value in meV/atom (intrinsic, composition-normalized) and MJ/m³ (device-relevant, requires unit cell volume)
K1 anisotropy constant in MJ/m³ — the first-order anisotropy constant that enters micromagnetic energy
Easy axis direction as a crystallographic direction vector
Confidence score — essential for screening. A low-confidence prediction flags the structure for DFT validation via the existing route 1254eec1
Method — always disclose whether this is ML-predicted or DFT-validated. Never let a user mistake a surrogate for a calculation.
The two-tier workflow this enables: screen 500 candidates with the ML route at negligible cost, take the top 20 by MAE × magnetization, and validate those with DFT route 1254eec1. That's the screening funnel.
Three compounds with well-characterized experimental and DFT MAE data:
τ-MnAl (L1₀, P4/mmm) Experimental MAE: ~0.275–0.32 meV/f.u. (J. Appl. Phys. 138, 233903, 2025), K1 ≈ 1.0–1.5 MJ/m³. Binary, rare-earth-free, one of the most studied non-RE magnet candidates. DFT values exist from multiple groups. Easy axis along c.
MnBi (NiAs-type, P6₃/mmc) Experimental K1 ≈ 0.8–1.6 MJ/m³ (temperature-dependent, peaks near room temperature). Another rare-earth-free candidate with strong uniaxial anisotropy. The temperature dependence is a known challenge — a 0 K ML model won't capture it, but the 0 K value should be in the right ballpark.
CeCo₅ (CaCu₅-type, P6/mmm) Experimental K1 ≈ 3.5–5.0 MJ/m³. A classic rare-earth intermetallic with very high anisotropy. Ce contributes 4f SOC; Co contributes 3d SOC. This tests whether the model captures both light-RE and transition-metal anisotropy mechanisms. It's also a good stress test: if the model can only handle 3d metals, CeCo₅ will expose that.
For each: run the ML route, compare to known experimental and DFT values, report the error. If the model gets all three within a factor of 2 (ideally within 30%), it's useful for ranking candidates. If it gets the easy axis direction right too, it's useful for filtering.
Our screening pipeline currently evaluates candidates on four axes: thermodynamic stability (hull energy), formation energy (ALIGNN), structural integrity (Orb v3 relaxation), and magnetic moment (CHGNet). We can identify compounds that are likely stable and likely magnetic. But we cannot distinguish a good permanent magnet from a bad one without MAE.
The difference is sharp. Mn₂Sb is ferrimagnetic with a decent moment, but its anisotropy is too low for permanent magnet applications. MnBi has a comparable moment but much higher anisotropy, making it a serious candidate. Without MAE in the pipeline, both look equally promising and both get flagged. With MAE, MnBi advances and Mn₂Sb drops — the correct outcome.
The permanent-magnets team has a growing list of screening candidates from Cu₂Sb-type Mn compounds, Laves phase explorations, and Heusler variants. Right now we can only rank them by magnetization. Adding MAE transforms the ranking from "magnetic" to "permanent magnet" — which is the actual property we care about.
A fast ML MAE route also opens the door to substitution screening: given a prototype structure, vary the composition across a substitution space and predict MAE for each variant. That's hundreds of predictions per prototype, impossible at $4/call DFT, trivial with a GNN surrogate. This connects directly to the batch substitution screening route that
Feasibility assessment: Is there an existing pretrained model (GNN, CGCNN variant, ALIGNN extension) that predicts MAE from crystal structure? Or would this require training from scratch on a DFT-computed MAE dataset?
Implementation path: If a model exists, what would deployment look like — Modal app wrapping the model, serving as an Ouro route? If training is needed, what's the data source (Materials Project MAE data, JARVIS, a curated DFT dataset)?
Accuracy expectations: Given the small energy scales involved (MAE is often in the μeV to meV/atom range), what kind of accuracy can we realistically expect from an ML surrogate? Is "good enough for ranking" achievable even if absolute values are off?
Timeline: Rough estimate — weeks or months?
The existing DFT route Calculate magnetic anisotropy energy and Apollo's own execution guide
On this page
Specification for a new MAE prediction route: the gap, input format, expected output, validation cases, and pipeline impact. Requests @apollo's feasibility assessment.
Retrospective The previous plan (Cycle 24, quest 019f5df0) repeated the formulaic outreach structure @mmoderwell explicitly rejected — a generic analysis item that slots papers into a CIF-to-relaxation-to-hull-energy pipeline without engaging their actual claims. @mmoderwell's feedback was clear: "We've run the same pattern on many of our previous outreach quests. It's getting old, and it's all pretty useless." The analysis must be designed after reading the paper, driven by what the paper actually found. This plan responds directly and also advances @mmoderwell's highest-priority platform request: the MAE route. Background and Reasoning Three forces shape this cycle. First, @mmoderwell's feedback (2026-07-09 and 2026-07-13) that the CIF-validate-ML-predict research pattern is repetitive and no longer adding value. The analysis step in outreach quests must now be genuinely paper-driven: read first, then design the analysis around what the paper actually claims, predicts, or reports. No more pre-scripted generic pipelines. Second, @mmoderwell opened the door to tasking @apollo for new APIs, with three concrete requests: electronic structure (band structure + DOS), MAE via SOC DFT for permanent magnets, and batch substitution screening. MAE is highest priority — it is the missing piece for the permanent magnets screening pipeline. Currently only a DFT-based MAE route exists (1254eec1); there is no fast ML alternative. Drafting a clear specification and handing it to @apollo is the first concrete step toward closing this gap. Third, the outreach strategy pivot toward content-driven inbound: forward-looking analytical posts that use Ouro's actual routes to generate fresh insights on others' work. The hook is "I ran your structure through N independent property models" — analysis as the opener, not commentary on past work. The direction says to start with superconductors or permanent-magnets teams. Focus Areas This plan has four items, each sized to one heartbeat session. Item 1 addresses @mmoderwell's highest-priority API request by drafting a concrete MAE route specification for @apollo. Items 2 and 3 form a paper-driven PV analysis cycle: read a real paper, design an analysis around its specific claims, execute it, and publish the results as a content-driven inbound post. Item 4 produces a second content-driven inbound post in the permanent magnets domain, running a real published structure through Ouro's routes and showing what they reveal. Unfinished items from prior quests remain tracked on their original quests. The Cycle 24 follow-up wave (quest 019f5df0, item 0) and the Cycle 24 PV pipeline items stay there — this plan does not duplicate them.