Activation-Prefactor Landscapes of Crystal Growth at Planar BCC and FCC Crystal-Melt Interfaces

We extend the local-structure-dependent (LSD) framework to planar crystal-melt interfaces (CMIs) of BCC Fe(100) and FCC Ni(100) to quantify how the local liquid structure governs solidification kinetics. Using machine-learning descriptors of atomic softness (S) and steady-state MD simulations, we decompose the kinetic factor into a superposition of Arrhenius channels, k(T,S) = k0(S) exp−ΔEa(S)/(kBT), and map the (ΔEa, k0) landscapes across S. Here, ΔEa(S) denotes the softness-resolved activation-energy barrier associated with thermally activated atomic attachment and rearrangement events at the crystal-melt interface within a given local-structure channel. The LSD model, parametrized only at low undercooling, accurately predicts growth rates over broader temperatures. Distinct structure-kinetics fingerprints emerge: in Fe(100), both ΔEa(S) and k0(S) decrease with increasing S (low-barrier + low-prefactor channels), whereas in Ni(100) they increase (high-barrier + high-prefactor channels), with rare low-S channels exhibiting apparent negative activation energies. These contrasting channel families rationalize the different temperature responses of BCC and FCC growth and reveal that kinetic anisotropy originates from the interplay between interface-induced ordering and local packing constraints. The orientation-resolved, softness-decomposed analysis provides a physically transparent route to “liquid-structure engineering”, and the results establish a transferable, data-driven basis for embedding softness statistics into mesoscale models to connect atomic mechanisms with microstructure control.