A battery cathode needs atoms that grip each other tightly but let lithium ions pass through. High K for structure, low K for transport. We screen materials by measuring both. The ones that score well on the coupling balance last longest.
K here is structural stability. Cathode materials with high K between lattice sites hold ions better. Low K = capacity fade. We screen for coupling.
A battery cathode has one job: hold lithium ions tightly when they’re supposed to stay, and let them go when they’re supposed to leave. That’s a coupling problem. Too tight and nothing flows. Too loose and the structure falls apart. The sweet spot is what makes a good battery.
We screened 66 cathode compositions in the nickel-manganese-cobalt space. 692,000 compositions per second. The question: what ratio gives the best tradeoff between capacity (how much energy) and stability (how many charge cycles)?
More nickel = more capacity but less stable. More manganese = more stable but less capacity. Cobalt sits in between. The Pareto frontier shows the optimal tradeoffs — and 17 of 28 optimal compositions still include cobalt. The “cobalt-free battery” dream is harder than the headlines suggest.
Same story for solar cells. We screened 1,352 perovskite compositions. FAPbI3 wins — closest to the optimal bandgap AND structurally stable. This matches what the entire solar industry already knows. The screening confirms it from coupling strength alone. No machine learning. No training data. Just the physics of how atoms hold together.
Honest limit: the stability weights are assumed, not measured. This is a fast filter, not a predictor. It eliminates obviously bad candidates and confirms known winners. It does not replace DFT calculations or experimental validation.
NMC (nickel-manganese-cobalt) cathodes are the dominant lithium-ion battery chemistry. The composition space is enormous: what ratio of Ni:Mn:Co gives the best tradeoff between capacity and structural stability?
K measures coupling strength between atoms in the crystal lattice. Higher K = more rigid structure = better cycle stability. But higher nickel = more capacity. The Pareto frontier reveals the optimal tradeoff.
| Composition | Capacity | Stability (K) | Cobalt |
|---|---|---|---|
| NMC 811 | 200 mAh/g | 0.472 | yes |
| NMC 622 | 175 mAh/g | 0.592 | yes |
| NMC 532 | 155 mAh/g | 0.624 | yes |
| NMC 111 | 150 mAh/g | 0.683 | yes |
10P correction: the original claim that Pareto-optimal compositions have zero cobalt was wrong. Full Pareto analysis shows 17 of 28 optimal compositions include cobalt. Under alternative stability weights (cobalt more stabilizing), 19/21 Pareto points include cobalt. The zero-cobalt result was an artifact of our Mn > Co > Ni stability weighting. Industry confirms: cobalt is being reduced but not eliminated (Samsung SDI NMC 622, Panasonic NCA both use cobalt). Our model is a fast filter, not a predictor. The stability weights are assumed, not measured.
ABX3 perovskites are the next generation of solar cells. The question: which combination of A-site cation, B-site metal, and X-site halide gives optimal bandgap for solar absorption?
Optimal bandgap is 1.34 eV (Shockley-Queisser limit). K measures the stability of the crystal structure — tolerance factor determines whether the perovskite forms at all.
| Composition | Bandgap (eV) | Tolerance | Stability (K) |
|---|---|---|---|
| FAPbI3 | 1.48 | 1.01 | 0.89 |
| MAPbI3 | 1.55 | 0.91 | 0.85 |
| CsPbI3 | 1.73 | 0.85 | 0.72 |
| FASnI3 | 1.41* | 1.05 | 0.00 |
FAPbI3 wins for single-junction cells — this matches industry consensus. But the result needs caveats:
10P correction: *Our bandgap formula gave FASnI3 = 1.85 eV. Literature value is 1.41 eV — actually CLOSER to the 1.34 eV optimum than FAPbI3. FASnI3 scores 0 because Sn2+ oxidizes in air (correct), but the bandgap we used was wrong. FAPbI3 wins for the right conclusion but partially the wrong reason: stability filtering saves us, not bandgap accuracy. Also missing: mixed compositions (FA0.95Cs0.05PbI3) hold actual efficiency records but weren't in our search space. Our screening only tested pure compositions.
Both screening engines use the same core: K (coupling strength) as a structural stability proxy. For cathodes, K comes from the crystal lattice coordination number and bond strength. For perovskites, K comes from the Goldschmidt tolerance factor and octahedral factor. No machine learning. No training data. Just the physics of how atoms couple.
All screening computed on Mac Mini M4, 35W. Code: pip install begump