Error Suppression and Mitigation — Overview

qc/mitigation qc/noise

Four categories of techniques exist for dealing with noise on today’s quantum hardware. They differ in when they act, what they cost, and whether they fix, flag, or statistically correct errors.

CategoryWhat it doesCost type
Error Correction (EC)Encodes qubits redundantly; syndromes detect and fix errors fault-tolerantlyExtra qubits (prohibitive near-term)
Error Detection (ED)Flags corrupted shots and discards them via post-selectionReduced usable sample size
Error Mitigation (EM)Corrects the statistics across many noisy runs so the expectation value estimate is closer to idealMore shots (time overhead)
Error Suppression (ES)Reshapes noise during execution (DD, PT)Modest circuit overhead

This section focuses on mitigation and suppression — the two categories Qiskit Runtime’s Estimator primitive implements.

The runtime default

At resilience_level=1 (default), Estimator already applies TREX — twirled readout error extinction. Everything else is opt-in.

Why whole-circuit switches aren’t enough

Qiskit Runtime exposes suppression and mitigation as circuit-wide policies — one setting applied uniformly to every gate or layer. This is sufficient for small workloads but has two gaps:

  1. No per-layer control — PEA and PEC internally use per-layer noise models, but you cannot inspect or override per-layer noise, or apply different strategies to different layers.
  2. No observable rewriting — you cannot absorb learned noise into the observable rather than the circuit (this is what PNA does).

These gaps are what Samplomatic and the Chapter 3 add-ons (PNA, SLC) fill.

Self-Check

  • Could you explain the difference between error correction, detection, mitigation, and suppression to someone new to this?
  • Why does this section focus on mitigation and suppression instead of correction?
  • What are the two specific gaps that whole-circuit switches leave open, and what fills them?