Predicted structural models for NaSbBr4 and LiSbBr4 from a prototype-transplant search with machine-learned potentials and DFT

Many ternary halides studied for solid-state batteries have no reported crystal structure. Their computed stability then depends on whichever structure is assumed, so the 3D crystal structure is the bottleneck. We transplanted 228 prototypes from the Alexandria database into AMX4 chemistries (A = Li, Na; M = Al, Ga, In, Sc, Y, Sb, Bi; X = Cl, Br, I; 9431 transplants) and screened them with the MACE-MPA-0 machine-learned potential. We then compared the best survivors with every database-listed polymorph and competing phase we computed, using PBE-D3. The main result is a proposed C2/c model, TP-NaSbBr4-01, predicted 6.2 meV/atom below NaBr + SbBr3 and 17.0 meV/atom below the lowest computed NaSbBr4 polymorph listed in Alexandria. The sign holds with D3(BJ), with the nonlocal vdW-DF-optB86b functional and without dispersion. Harmonic phonons at 2x2x2 sampling show no imaginary mode, which does not by itself establish dynamical stability. A secondary candidate with less evidence, TP-LiSbBr4-01, lies below its competing phases with D3 zero damping, D3(BJ) and optB86b (7.7, 8.4 and 10.9 meV/atom). With plain PBE, it lies +2.05 meV/atom above them. Its phonons at 2x2x2 sampling also show no imaginary mode. The method's own errors are as large as or larger than these margins. PBE-D3 mis-orders the SbBr3 polymorphs by +15.2 meV/atom, and optB86b mis-orders them by +21.6 meV/atom (with PBE-generated PAW datasets), so the error is not a D3 artifact. The protocol also fails its one false-positive control. It ranks a hypothetical LiAlBr4 model 4.5 meV/atom below a LiAlBr4 cell built on the LiAlCl4 structure type. At this level of theory, both models are candidates for diffraction testing. Their position below the computed binaries does not establish them as phases. No electrolyte suitability is shown. Nothing was synthesized. Machine-learned molecular dynamics found no long-range ion diffusion in the pristine crystals. Sb(III) is expected to be reduced at Li and Na anode potentials. The computed 0 K PBE-D3 windows are narrow (2.47-3.35 V vs Na/Na+, 2.38-3.27 V vs Li/Li+). They lie above the metal potential and below common cathodes, and the oxidation limit is an upper bound. The PBE gaps (2.5 and 2.4 eV) are underestimates. Models in five other screened compositions, including NaBiCl4, also meet all four gates (one machine-learned, three DFT). They did not receive the dispersion, phonon and window checks, so they are reported briefly and not pursued.