Quantum Benchmark Zoo

← All benchmarks

Direct fidelity estimation

Characterization

Estimates how close a lab state or gate is to its ideal pure target from a few importance-sampled Pauli measurements, sidestepping the exponential cost of full tomography.

Measures
Fidelity between an implemented state or process and a pure target state or unitary, to constant additive error, from a few random Pauli expectation values
Introduced by
Flammia & Liu, da Silva, Landon-Cardinal & Poulin (2011)
Also known as
DFE, Monte Carlo quantum process certification

Direct fidelity estimation (DFE) answers the certification question (how close is my state or gate to the one I meant to make?) without paying for full quantum state tomography. It was proposed twice, independently, in April 2011: da Silva, Landon-Cardinal & Poulin posted first (PRL 107, 210404), and the name comes from Flammia & Liu (PRL 106, 230501); cite both. The da Silva formulation is also known as Monte Carlo (process) certification, after the 2012 Steffen et al. experiment.

How it works

The fidelity to a pure target expands as a sum over Pauli operators of the target’s Pauli coefficients times the lab state’s. DFE samples a small number of Pauli operators with probability proportional to the target’s squared coefficients (importance sampling), estimates each sampled Pauli expectation value on hardware, and averages, giving the fidelity to constant additive error. The number of measurement settings is constant in system size. Total shot cost depends on the target: stabilizer states such as GHZ states are especially cheap. Gates are certified the same way through the entanglement fidelity of the corresponding channel.

Strengths and limitations

For certification, DFE is exponentially cheaper than tomography, and it remains a standard primitive 15 years on, with implementations shipping in Cirq and Rigetti’s forest-benchmarking. But it returns one number for a user-chosen target, with no fixed workload and no standardized device score, so it is measurement machinery for fidelity-based benchmarks rather than a benchmark itself, and it offers no diagnosis of what went wrong. The original protocols assume a pure target state or unitary (a 2024 generalization extends DFE to generic mixed states), and shot cost is target- and precision-dependent, with tightened resource bounds still an active topic in 2026. Beware a naming collision: Superstaq’s “DFE” tool implements the randomized-measurement cross-platform verification protocol of Elben et al., a distinct method sharing the acronym; classical shadows offer yet another route to fidelity estimation.

Key papers

Reference implementations

  • Quantum state tomography : The baseline characterization protocol: reconstructs the full density matrix of a prepared state from repeated measurements in an informationally complete set of bases.
  • Cross-platform verification : Randomized-measurement protocol that estimates the fidelity between quantum states prepared on two different devices using only classically communicated random unitaries and measurement outcomes
  • GHZ State Fidelity : Whole-device entanglement test that prepares an N-qubit GHZ state and certifies genuine multipartite entanglement whenever the measured fidelity clears 0.5.