API Reference

Top-level exports 

from kineticEQ import Config, Engine, run, Result, BGK1D, BGK2D2V

Config 

Config(
    model: str = "BGK1D1V",
    scheme: str = "explicit",
    backend: str = "torch",
    device: str = "cuda",
    dtype: str = "float64",
    log_level: str = "info",
    model_cfg: Any | None = None,
    use_tqdm: str = "true",
)

Core fields

  • model: BGK1D1V / BGK2D2V
  • scheme: explicit / implicit / holo
  • backend: torch / cuda_kernel / cpu_kernel
  • device: cuda / cpu / mps
  • dtype: float32 / float64
  • model_cfg: model-specific dataclass
  • use_tqdm: true / false

Aliases

  • model: bgk1d, bgk1d1v, bgk2d2v
  • scheme: exp, imp, hl
  • backend: pytorch, cuda_backend, cpu_backend
  • dtype: fp32, fp64

Convenience accessors

  • model_name
  • scheme_name
  • backend_name
  • dtype_name
  • log_level_name
  • use_tqdm_name
  • use_tqdm_bool
  • as_dict

Engine 

Engine(config: Config, apply_logging_flag: bool = True)

Responsibilities

  • fill or validate model_cfg
  • fill default scheme_params
  • apply logging
  • validate the requested device
  • build the state
  • build the registered stepper

run 

engine.run() -> Result

The runtime loop advances model_cfg.time.n_steps iterations. If the stepper exposes a benchlog attribute, it is periodically written to the progress output.

run 

run(config: Config) -> Result

This is a thin shortcut for Engine(config).run().

Result 

@dataclass
class Result:
    metrics: dict[str, float] | None = None
    payload: dict[str, Any] | None = None

BGK1D dataclasses

BGK1D.ModelConfig 

BGK1D.ModelConfig(
    grid: Grid1D1V = Grid1D1V(),
    time: TimeConfig = TimeConfig(),
    params: BGK1D1VParams = BGK1D1VParams(),
    initial: InitialCondition1D = InitialCondition1D(),
    scheme_params: Any = None,
)

BGK1D.Grid1D1V

  • nx: int = 124
  • nv: int = 64
  • Lx: float = 1.0
  • v_max: float = 10.0

BGK1D.TimeConfig

  • dt: float = 5e-4
  • T_total: float = 0.05
  • n_steps = ceil(T_total / dt)

BGK1D.BGK1D1VParams

  • tau_tilde: float = 0.5

BGK1D.InitialCondition1D

  • initial_regions: tuple[Any, ...]
  • each element provides x_range, n, u, and T

BGK1D.implicit.Params

Contains Picard tolerances, convergence mode, Anderson-acceleration settings, and CNN warm-start options.

BGK1D.holo.Params

Contains HO/LO iteration counts and tolerances together with consistency flags.

BGK2D2V dataclasses

BGK2D2V.Grid2D2V

  • nx: int = 124
  • ny: int = 124
  • nx_v: int = 16
  • ny_v: int = 16
  • Lx: float = 1.0
  • Ly: float = 1.0
  • v_max: float = 10.0

BGK2D2V.TimeConfig

  • dt: float = 5e-3
  • T_total: float = 0.05
  • n_steps = int(T_total / dt)

BGK2D2V.BGK2D2VParams

  • tau_tilde: float = 0.5

Note:

  • the current BGK2D2V.ModelConfig does not define scheme_params, so the Engine path is not operational.