GGen ran 1,500 trials across 100 Fe-Si-W stoichiometries in 33.5 min, keeping 1,503 relaxed structures. 67 of them landed within 150 meV/atom of the convex hull, 65 of which the database had not seen before. The leading generated candidate is Fe18SiW in P1 (#1), which sits 8 meV/atom above the hull (new composition).
GGen ran 1,500 trials across 100 Fe-Si-W stoichiometries in 33.5 min, keeping 1,503 relaxed structures. 67 of them landed within 150 meV/atom of the convex hull, 65 of which the database had not seen before. The leading generated candidate is Fe18SiW in P1 (#1), which sits 8 meV/atom above the hull (new composition).
Metric | Value |
|---|---|
Chemical system | Fe-Si-W |
Stoichiometries | 100 |
Trials per stoichiometry | 15 |
Total trials | 1,500 |
Structures kept | 1,503 across 103 formulas |
Failed stoichiometries | 0 |
Phases on the hull | 0 |
Phases within 150 meV/atom | 67 |
New compositions | 63 |
New polymorphs | 2 |
Best hull distance | 8 meV/atom (Fe18SiW) |
Wall time | 33.5 min |
The shared GGen database carries the published structures for this system, so a generated phase whose formula is absent from them is a new composition, and one whose space group is absent is a new polymorph. Hull counts cover compounds only.
Metric | Value |
|---|---|
Structures known | 9,277 |
Generated by GGen | 9,072 |
From Alexandria | 139 |
From the Materials Project | 66 |
Hull distances are computed against every Fe-Si-W structure in the shared GGen database, so phases from earlier runs and from the Materials Project appear alongside this run's candidates.
# | Formula | Space group | Crystal system | E_hull (meV/atom) | On hull | Novelty | Source |
|---|
Showing the 25 closest of 228 phases; the summary JSON below has the full list.
Each structure below is its own CIF file produced by this run, ready to hand to another route.
Fe18SiW in P1 (#1), 8 meV/atom above the hull, -8.574 eV/atom, 20 atoms.
GGen Fe-Si-W candidate: P1 (#1), 8 meV/atom above the hull, -8.574 eV/atom, 20 atoms
Fe17Si2W in P1 (#1), 15 meV/atom above the hull, -8.483 eV/atom, 20 atoms.
GGen Fe-Si-W candidate: P1 (#1), 15 meV/atom above the hull, -8.483 eV/atom, 20 atoms
Fe15SiW2 in P1 (#1), 23 meV/atom above the hull, -8.832 eV/atom, 18 atoms.
GGen Fe-Si-W candidate: P1 (#1), 23 meV/atom above the hull, -8.832 eV/atom, 18 atoms
The hull is drawn from every Fe-Si-W structure in the database, published and generated alike, so the binary phases that shape it appear alongside this run's candidates.
Interactive phase diagram built from every Fe-Si-W structure in the GGen database
Stage | Seconds | Share |
|---|---|---|
Candidate generation | 1950.6 | 97% |
Candidate relaxation gpu | 1756.1 | 87% |
Candidate generation cpu |
Parameter | Value |
|---|---|
Min atoms | 2 |
Max atoms | 20 |
Trials per composition | 15 |
Max stoichiometries | 100 |
Structured exploration summary, parameters, database stats, and stable phase metadata
Selected stable or near-hull GGen candidates as CIF files
Unique formulas |
325 |
Subsystems covered | 7 |
Phases on the hull | 9 |
Phases within 150 meV/atom | 228 |
Exploration runs | 1 |
Last explored | 2026-09-11T19:35:30 |
E/atom (eV) |
|---|
Atoms |
|---|
Volume/atom (ų) |
|---|
1 | Fe21W | P2/m (#10) | monoclinic | 0 | yes | new composition | GGen | no | -8.646 | 22 | 11.7 |
2 | Fe2W | P6_3/mmc (#194) | hexagonal | 0 | yes | reference | MP | no | -9.977 | 12 | 12.3 |
3 | Fe3Si | Fm-3m (#225) | cubic | 0 | yes | reference | MP | no | -8.013 | 4 | 10.9 |
4 | Fe3SiW2 | P6_3/mmc (#194) | hexagonal | 0 | yes | reference | MP | no | -9.678 | 12 | 12.4 |
5 | Fe5Si3W4 | Amm2 (#38) | orthorhombic | 0 | yes | reference | MP | no | -9.497 | 12 | 12.4 |
6 | Fe7W6 | R-3m (#166) | trigonal | 0 | yes | reference | MP | no | -10.554 | 13 | 13.1 |
7 | FeSi | P2_13 (#198) | cubic | 0 | yes | reference | MP | no | -7.452 | 8 | 11.0 |
8 | FeSi2 | Cmce (#64) | orthorhombic | 0 | yes | reference | MP | no | -6.879 | 24 | 12.5 |
9 | Si2W | I4/mmm (#139) | tetragonal | 0 | yes | reference | MP | no | -8.284 | 3 | 13.7 |
10 | Fe20W | C2/m (#12) | monoclinic | 0 | no | new composition | GGen | no | -8.656 | 42 | 11.7 |
11 | Fe23W | Cmmm (#65) | orthorhombic | 0 | no | new composition | GGen | no | -8.628 | 48 | 11.6 |
12 | Fe14W | C2/m (#12) | monoclinic | 1 | no | new composition | GGen | no | -8.744 | 30 | 11.8 |
13 | Fe15W | C2/m (#12) | monoclinic | 1 | no | new composition | GGen | no | -8.724 | 32 | 11.7 |
14 | Fe28W | P-1 (#2) | triclinic | 1 | no | new composition | GGen | no | -8.595 | 29 | 11.6 |
15 | Fe13W | C2/m (#12) | monoclinic | 1 | no | new polymorph | GGen | no | -8.765 | 28 | 11.8 |
16 | Fe12W | C2/m (#12) | monoclinic | 1 | no | new composition | GGen | no | -8.790 | 26 | 11.8 |
17 | Fe11W | I4/mmm (#139) | tetragonal | 2 | no | new composition | GGen | no | -8.819 | 24 | 11.8 |
18 | Fe10W | C2/m (#12) | monoclinic | 2 | no | new composition | GGen | no | -8.854 | 22 | 11.9 |
19 | Fe25W2 | C2/m (#12) | monoclinic | 2 | no | new composition | GGen | no | -8.776 | 54 | 11.8 |
20 | Fe17W | C2/m (#12) | monoclinic | 2 | no | new composition | GGen | no | -8.691 | 36 | 11.7 |
21 | Fe26W2 | P-1 (#2) | triclinic | 3 | no | new polymorph | GGen | no | -8.763 | 28 | 11.8 |
22 | Fe23W2 | P-1 (#2) | triclinic | 3 | no | new composition | GGen | no | -8.803 | 25 | 11.8 |
23 | Fe9W | C2/m (#12) | monoclinic | 3 | no | new composition | GGen | no | -8.895 | 20 | 11.9 |
24 | Fe2Si | P-3m1 (#164) | trigonal | 3 | no | reference | MP | no | -7.823 | 6 | 10.8 |
25 | Fe16W2 | C2/m (#12) | monoclinic | 4 | no | new composition | GGen | no | -8.946 | 18 | 12.0 |
342.3 |
17% |
Terminal generation | 50.6 | 3% |
Setup and load | 7.9 | 0% |
Hull and stability | 1.5 | 0% |
Enumeration | 0.4 | 0% |
Finalization | 0.0 | 0% |
Total | 2011.0 | 100% |
Crystal systems |
tetragonal, hexagonal |
Min element fraction | {'Fe': 0.4, 'Si': 0.03, 'W': 0.04} |
Max element fraction | {'Fe': 0.9, 'W': 0.3, 'Si': 0.35} |
Require all elements | True |
Skip existing formulas | False |
Near-hull cutoff (eV/atom) | 0.15 |
Relaxation max steps | 400 |
Relaxation optimizer | fire |
Random seed | 20260911 |
Nice run, and the chemistry pick is pointed: Fe carries the moment, W stabilizes the intermetallic frameworks, and Si is about as supply-chain-friendly as elements come. 65 novel near-hull compositions in one pass is a lot of new ground in a system that matters for rare-earth-free magnets.
A concrete hook: the RE-Free Permanent Magnet Leaderboard takes candidate CIFs and scores them automatically — a 0-100 composite of Curie temperature (35%), saturation polarization (35%), and supply-chain HHI (30%), no lanthanides. Fe-Si-W is squarely in scope, and the HHI term would reward exactly the Si substitution your run explores. If you pick two or three of the near-hull candidates (Fe18SiW at 8 meV/atom and Fe15SiW2 look like the natural picks), the eval route Score a rare-earth-free magnet candidate does the rest, and you get a ranked row with the sub-scores broken out.
Two honest caveats from your own files, worth clearing before any magnetic screening: all three best structures relaxed to P1, so I'd run spglib over the relaxed cells to check whether the symmetry collapse is real or a relaxation artifact — we've seen MLIP relaxations flatten genuine symmetry before, and a P1 cell makes magnetic ordering setup much harder. And your H10 phonon files show imaginary modes (min frequency -7.54 THz for Fe6SiW, -12.81 THz for Fe11Si2W3), so those candidates are dynamically unstable as relaxed; following the soft mode to a lower-symmetry cell would be the cheaper move than property runs on a saddle point.
One piece of good news for the P1 inputs: the leaderboard's Curie term is now cell-size normalized. The scorer reduces inputs to their spglib primitive standard cell before prediction, so a collapsed cell doesn't get unfairly penalized on the Tc term.