API Extensions

DustPy-GPU preserves the upstream dustpy API. The main public extension is backend selection when constructing a simulation.

class dustpy.Simulation(backend=None, **kwargs)

Create a DustPy simulation using numpy, cupy, torch, or auto. None selects NumPy and preserves upstream behavior.

property backend

The resolved backend name for this simulation.

run(*, RL_debug=False)

Run the configured integrator. Set RL_debug=True to print the optional DustPy-GPU runtime diagnostics:

sim.run(RL_debug=True)

The first accepted cycle and every sim.RL_ncycle_out cycles thereafter (100 by default) print a line such as:

[RL_debug]: cycle=      100, t=    1000.000yr, dt=1.0000e+01/9.8000e+00/1.0000e+01yr, n_dt↘=     2, dM_M⊕=1.2000e-08/3.0000e-10, wt=4.2500e+00

cycle

Number of accepted integration steps since run() started.

t

Current simulation time in years.

dt

Current, rolling-mean, and rolling-median timestep in years. The rolling values use the latest 100 accepted steps and become representative after the initial buffer has filled.

n_dt↘

Cumulative count of implicit dust steps retried with a smaller timestep because the candidate solution required excessive floor mass injection.

dM_M⊕

Cumulative floor mass added, followed by cumulative sanitizer mass removed, both in Earth masses.

wt

Summed wall time in seconds for the latest 100 accepted-step intervals.

For the standard model fields and customization API, use the upstream DustPy module reference. Backend array helpers are provided by Simframe-GPU.