# Starting from a saved interior field A saved field can be the input to a new calculation beginning at its physical position inside a guide. Use this to compare response models over the same subsequent interval. A new segment records a new launch; continuing an interrupted calculation is instead described under [file continuation](run-access.md#writing-to-a-file-and-resuming-it). ## Starting from an interior saved plane An explicit modal spectrum can be launched at `initial_z`, in metres, while retaining the original guide length, absolute density and pressure profiles, propagation frame and coarse grid. The default is zero. A positive launch requires `noise=None`, no fixed schedule, and an array shaped `(n_modes, n_pol, n_omega)` in the internal FFT convention with true-power amplitudes. Every requested save must lie strictly beyond the launch, and the launch must precede the guide end. To retain an existing simulation's translated tables, use `with_input`: ```{code-block} python from solprop.output import open_file with open_file("parent.h5") as parent_output: plane_index = 10 z_start = float(parent_output["z"][plane_index]) spectrum = parent_output.fields(convention="fft")[plane_index] parent_hash = parent_output.metadata.configuration_hash() segment = simulation.with_input(spectrum, initial_z=z_start, noise=None) segment.run("segment.h5") ``` Here `simulation` is the matching parent configuration. An explicit `initial_z` filters its existing save planes to those strictly beyond the launch; omitting the keyword keeps the existing launch and saves. The new output includes the launch plane, and its metadata hashes the new input and launch coordinate. Normalisation uses the effective index at `z_start`, so a pressure gradient is sampled at the same physical position as the saved parent field. To change the response model, construct a new `Simulation` with the parent grid, modes, profiles and frame, the replacement responses, this `spectrum` and `initial_z=z_start`; include the parent hash and plane index in its `configuration` provenance and write a new output file. The same continuation can be made through plane exchange, which checks the basis and records the provenance for you: `couple(field_plane(parent_output, plane_index), simulation.input_port(z_start), model=CouplingModel("same_basis")).launch` is a launch that `simulation.with_input` accepts, at `z_start`, with the parent named in `configuration["coupling"]` ([coupling simulations](../workflows/coupling.md#relaunching-into-the-same-basis)). For a TDSE response, `SimulationSettings.tdse_execution` explicitly selects the atomic runtime resources beside `backend="numpy_ref"` and a `Multistep` integrator. The declared response sampling refines the existing fine lattice once, retaining the coarse grid, period and collars. An inherited standard temporal collar is evaluated on the finer lattice with the same physical reference length; a custom absorber must already carry matching fine samples. Changing execution device, batch width or memory limits does not alter the scientific resume identity.