Target specifications, execution capabilities and machine resources¶
A Target sets geometry, waveform, control and sampling constraints. A Backend provides the submission and execution path. Simulation methods and machine resources determine which numerical calculations are possible. Passing one check does not satisfy all three.
A target’s atom limit, for example, does not guarantee that a laptop has enough memory for a density matrix of that size. Likewise, local simulation of a control sequence does not show that a physical device can implement it.
Read capability status¶
A capability can be supported, representable only in the intermediate representation, or unsupported. ir_only allows semantics to be stored and checked; it does not enable execution. Validation also checks parameter ranges, timing grids, local addressing and channel combinations, rather than just feature names.
MockNeutralAtomTarget provides deterministic local specifications. Its v0_1(), v0_2(), local_ahs_v0_1() and control_vocabulary_v0_1() factories serve different purposes. In the control-vocabulary target, DMM is ir_only and Raman is unsupported. Do not transfer capabilities from one factory to another by assumption.
None of these public targets provides live hardware or cloud submission. Target identifiers, snapshot timestamps, reference compiler records and local results do not replace calibration or execution records from a real device.
What to retain with an experiment¶
Use the same target specifications for validation and compilation. Keep the snapshot and references so a changed program is not interpreted using stale assumptions. When a request is rejected, follow the diagnostics to the unsupported control or exceeded range. For a resource-planning failure, check state representation, memory estimates and concurrency in the resource guide.
See reference compilation for the workflow and current limitations for the broader support boundary.