Advanced techniques with Qedma QESEM

Use the following techniques to get the most out of Qedma QESEM. For every option and its default value, see the API reference.

---

Estimate QPU time first

Set the estimate_time_only option before you run a full mitigation job:

• Use "analytical" for a quick upper bound that consumes no QPU time. It is useful for comparing circuits and precisions.
• Use "empirical" for a more accurate estimate. It runs a small number of circuits, and its partial mitigation results might already be enough for small circuits or low target precision.

Run the mitigation job on the same backend that you used for the estimate, because the estimate changes from one backend to another.

---

Balance precision and QPU time

QPU time grows roughly with the inverse square of the target precision. Set default_precision for all PUBs, or set a precision on an individual PUB. Use max_execution_time (in seconds) to cap the QPU time of the whole QESEM process.

---

Choose a transpilation level

• "standard" (default) prepares several transpilations and picks the one that minimizes QPU time.
• "minimal_with_layout_opt" keeps your circuit structure and respects barriers. Transpile your circuits to the backend basis gates first. Use this level for parameterized circuits and multi-PUB jobs, which let QESEM share characterization across circuit instances.

---

Choose an execution mode

Set execution_mode to "batch" (default) for lower QPU usage, or to "session" for a shorter time to result. Keep parallel_execution enabled so that QESEM can run your circuit on several high-fidelity patches of the QPU at once.

---

Request noise-scaled results

To get expectation values at selected noise scale factors, pass a dictionary that maps each scale factor to its target precision as the PUB precision. This is called Quasi-probabilistic Error Tuning (QET).