Read and plot results

Simulation.run() returns a Result. This guide shows the common ways to get numbers and pictures out of it. Field and flux data are PyTorch tensors on the scene device; call .detach().cpu().numpy() when you need NumPy. Plane-monitor coordinates (x, y, z) are NumPy arrays. Every member is listed on the Result API page.

Read a monitor#

result.monitor(name) returns a dictionary. The keys depend on the monitor type; the Monitors API lists them per class.

probe = result.monitor("probe")          # PointMonitor
ez = probe["data"]                        # complex phasor of the first requested component
ez_again = probe["Ez"]                    # the same, by component name
 
plane = result.monitor("field_xz")       # PlaneMonitor on axis="y"
ez_plane = plane["Ez"]                    # complex tensor, shape (Nx, Nz)
x, z = plane["x"], plane["z"]             # sample coordinates, m (NumPy arrays)
 
flux = result.monitor("flux_up")         # FluxMonitor
power = flux["flux"]                      # real power through the plane

Plane monitors span the whole grid plane, including PML cells; use the coordinate tensors to crop to the physical domain.

Pick a frequency#

With several frequencies, monitor tensors carry a leading frequency axis. result.at(...) returns a view fixed to one frequency:

view = result.at(frequency=1.5e9)         # or result.at(freq_index=0)
ez_plane = view.monitor("field_xz")["Ez"]

The frequency must match one of result.frequencies (relative tolerance 1e-9).

Full-grid fields and materials#

Run with full_field_dft=True to keep the electric-field spectrum on the whole grid:

sim = mw.Simulation.fdtd(scene, frequencies=[1.5e9], full_field_dft=True)
result = sim.run()
ex, ey, ez = result.E.x, result.E.y, result.E.z
eps = result.material("eps_r")            # compiled relative permittivity

FDTD and FDFD results store the electric field only; result.H components raise KeyError. Record magnetic fields with an H-only PlaneMonitor or PointMonitor instead.

Plot#

import matplotlib.pyplot as plt
 
result.plot.field(axis="y", position=0.0, component="Ez")   # needs full_field_dft=True
plt.gcf().savefig("field.png", dpi=120, bbox_inches="tight")
 
result.plot.material("eps_r", axis="z", position=0.0)

plot.field draws the field magnitude and the permittivity side by side and calls plt.show(); it returns None, so save the current figure with plt.gcf(). component accepts "abs" (the default) or a component name. For custom figures, plot monitor tensors with Matplotlib directly.

Run statistics#

result.stats() returns a dictionary about the run:

stats = result.stats()
print(stats["steps_run"], stats["dt"], stats["elapsed_s"])
print(stats["time_step"]["limiting_factor"])   # "frequency_resolution" or "material_cfl"
print(stats["cuda_graph_active"])

Use it to check the time step, whether early shutoff triggered (shutoff_triggered), the CPML memory mode, and throughput (ms_per_step).

Save and reload#

result.save("run.pt")
restored = mw.Result.load("run.pt", scene=scene)
print(restored.monitor("flux_up")["flux"])

save writes a detached CPU snapshot of fields, monitors, ports, circuits, networks and statistics. The scene is not saved, so pass it back to load. Snapshot files use pickle; load them only from sources you trust. For large multi-GPU results, save_sharded and load_sharded write one file per shard.

Engineering quantities#

Related#

• Result API
• Monitors API