{
  "field_id": "F039",
  "field": "computational fluid dynamics",
  "directions": [
    {
      "direction_id": "G0",
      "name": "Flow-informed geometry design",
      "scope": "Use CFD to evaluate or guide design search, selecting and optimizing the persistent geometry of shapes, channels, and fluid devices to improve flow-related performance."
    },
    {
      "direction_id": "G1",
      "name": "Learned flow-result prediction and generation",
      "scope": "Learn to produce flow fields, flow-state evolution, or flow-performance results to replace expensive CFD result acquisition, or to characterize the distribution of variation in plausible flow states; encompasses both rapid prediction for specified inputs and conditional or unconditional sampling of flow states."
    },
    {
      "direction_id": "G2",
      "name": "Flow reconstruction from incomplete observations",
      "scope": "Infer unobserved fluid motion and field information from sparse, missing, coarse-grained, or visual observations."
    },
    {
      "direction_id": "G3",
      "name": "Turbulence closure modeling and assessment",
      "scope": "Develop, update, or calibrate in-solver closure representations for flow physics that cannot be directly resolved, and assess associated model error and uncertainty; includes subgrid-scale and near-wall closures in coarse-scale simulations, as well as RANS closure calibration and probabilistic assessment."
    },
    {
      "direction_id": "G4",
      "name": "Learned spatial operations for CFD solvers",
      "scope": "Improve meshes or local spatial representations and decisions to support accurate, stable numerical solutions while a CFD solver continues to advance the computation."
    }
  ]
}