{
  "field_id": "F038",
  "field": "computational chemistry",
  "directions": [
    {
      "direction_id": "G0",
      "name": "Learning for electronic-structure calculations",
      "scope": "Use learning methods to solve or improve electronic-structure calculations, including directly optimizing neural quantum states and energy, and learning Hamiltonians, charge densities, or density-functional approximations for use in subsequent calculations."
    },
    {
      "direction_id": "G1",
      "name": "Learned interaction potentials for simulation",
      "scope": "Develop learned interaction energies and forces that govern atomistic or coarse-grained simulations, and study the training, transfer, and simulation reliability of the resulting potentials."
    },
    {
      "direction_id": "G2",
      "name": "Chemical representation and observable prediction",
      "scope": "Learn reusable representations from information about known molecules, materials, mixtures, or candidate–target pairs, or predict their properties, interactions, and observable signals; includes physics-guided prediction of fluid thermodynamic properties."
    },
    {
      "direction_id": "G3",
      "name": "Discovery and design of new chemical candidates",
      "scope": "Generate or search for new molecular, crystal, and framework-material candidates. Outcomes may include plausible new molecular identities and connectivity, candidates meeting design objectives, binders for specified targets, or new material compositions and structures."
    },
    {
      "direction_id": "G4",
      "name": "Molecular three-dimensional configuration generation and search",
      "scope": "Obtain plausible molecular spatial configurations through three-dimensional molecular construction, conformer sampling, or binding-pose search, with the spatial configuration itself as the central outcome."
    },
    {
      "direction_id": "G5",
      "name": "Molecular ensemble sampling and kinetic reduction",
      "scope": "Address statistical and dynamical information that is difficult to obtain directly in high-dimensional molecular configuration spaces: estimate equilibrium distributions, integrals, and free energies, or identify transition paths and their low-dimensional kinetic coordinates."
    },
    {
      "direction_id": "G6",
      "name": "Single-step chemical transformation inference",
      "scope": "Infer transformation outcomes, products, or possible reactants for an individual chemical reaction step, while accounting for multiple plausible reaction modes in single-step prediction."
    }
  ]
}