Synthesis Planner
Rank solid-state synthesis routes to a target phase from simulated convex-hull data. Every
precursor set is balanced (with optional gas release or uptake from the atmosphere), scored
by how cleanly the target out-competes the other phases that can form from the same mixture
(driving forces, inverse hull energy and pairwise interfaces after Chen et al., Nat. Synth. 2024), and turned into a bench recipe. The same plan_synthesis() function ships as
an LLM tool definition (SYNTHESIS_PLANNER_TOOL); Copy summary shows what
an agent sees.
BaCO3 + TiO2 in air, with Ba2TiO4 as the known intermediate
Provide thermodynamic entries and a target to plan a synthesis.
How the ranking works
- Reaction energy per atom of target at the chosen temperature: solids keep their 0 K computed energies, open gases get μ(T, p) = Hf − T·S + kT ln(p/p°), so carbonate decomposition and O2 release turn on with temperature. The onset is where a gas-releasing reaction becomes downhill.
- Competing phases: every near-hull phase that the same precursor mixture can form, with its driving force per atom of reacting mixture (pymatgen InterfacialReactivity convention). The selectivity margin is the target's driving force minus the strongest competitor's; negative means the target forms first.
- Inverse hull energy: how far the target sits below the hull of all other reachable phases at its own composition; larger means intermediates are less likely to persist.
- Practicality from a library of commercial precursors (hygroscopic, air-sensitive, hazards, decomposition and melting points) that also feeds the heating window of the recipe.