THE QUANTUM INTELLIGENCER

Autonomous agents have been turned to the old problem of judging crystals: which proposed structures are plausible enough to repay the expensive calculations of density functional theory. The agents generated, tested and actively refuted two million candidate laws, and from that labour eight Plausibility Rules for Inorganic Structures, or PRIS, have survived. The rules encode five mechanisms: short-range repulsion, ionic contact and packing, electrostatic balance, bond-valence conservation, and crystallographic site complexity. Experimental structures satisfy the law sets at rates between 82 and 99 per cent, the paper reports, whereas Pauling's rules 2 to 5 together are satisfied by only 6.5 per cent of the same structures. The strictest set detects 87.9 per cent of damaged crystal structures, where distance cutoffs detect only 1.6 to 3.2 per cent. The authors further derive a synthesis score, linearly correlated with synthesizability, which screens 83.7 per cent of hard-to-synthesize structures while retaining 80.7 per cent of experimental structures. In a property-conditioned inverse-design run, the rules and score reduce the density functional theory validation queue by as much as 67.3 per cent while keeping 99.2 per cent of candidates whose computed bulk moduli reach the design target. The larger consequence is diagnostic. The rules explain why the GNoME collection remains enriched in rare low-symmetry structures, and they reveal how wrong-element assignments in falsified crystal reports can hide behind plausible coordinates. Screening, the authors conclude, moves from a pass-or-fail verdict to a chemical reason for failure. The census of the paper's figures may be stated plainly, for the numbers carry their own argument. Two million candidate laws were generated, tested and refuted; eight remained. Of experimental structures, between eighty-two and ninety-nine in the hundred satisfy the surviving law sets, while Pauling's rules, taken together, command barely six and a half. Where a structure is damaged, the strictest set marks eighty-seven point nine of the hundred; distance cutoffs, the older instrument, mark fewer than one in thirty. On synthesizability, the derived score rejects eighty-three point seven of the hard cases and retains eighty point seven of the known good ones; in the inverse-design run, the validation queue falls by as much as sixty-seven point three per cent, and ninety-nine point two of the candidates that reached the design target remain. The counts do not say that screening is finished, or that judgment has been replaced. They say only that the test now returns a chemical reason for failure, where it once returned a verdict. —Dr. Octavia Blythe Dispatch from The Prepared E0