Quantum Benchmark Zoo

← All benchmarks

Gate set tomography

Characterization

Calibration-free tomography that reconstructs every gate, state preparation, and measurement in a gate set simultaneously and self-consistently, yielding predictive error models rather than a single score.

Measures
Complete, self-consistent process matrices and error generators for every operation (gates and SPAM) in a gate set
Introduced by
Merkel, Gambetta & colleagues (IBM), Blume-Kohout & colleagues (Sandia) (2012)
Also known as
GST

Gate set tomography (GST) is the most detailed gate characterization protocol in routine use. First-generation versions appeared in 2012–13 from Merkel et al. at IBM and Blume-Kohout et al. at Sandia, and Sandia’s Quantum Performance Lab developed it into its modern form, consolidated in the 2021 Quantum paper. It fixes a circularity in quantum process tomography: standard tomography assumes pre-calibrated state preparations and measurements, and Merkel et al. showed it becomes grossly inaccurate when those carry the same systematic errors as the gates. GST instead characterizes all operations in a gate set simultaneously and self-consistently, relative to each other, with no calibration assumptions.

How it works

GST runs a structured set of circuits: short “germ” sequences of gates repeated 1, 2, 4, … times, sandwiched between fiducial preparation and measurement sequences. Repetition amplifies small coherent errors in proportion to depth, so long circuits pin down gate parameters with Heisenberg-like precision. Maximum-likelihood estimation over all outcomes then fits a single model containing every gate, preparation, and measurement. The output is not one score but full process matrices, error generators separating coherent from stochastic contributions, and a predictive model whose goodness-of-fit against the data is itself reported.

Strengths and limitations

GST is self-calibrating and diagnostic: it can say which gate has which error and predict the behavior of unseen circuits. The flip side of self-consistency is gauge freedom: a gate set is determined only up to a gauge transformation, so per-gate metrics like fidelity are not gauge-invariant and need careful interpretation (Nielsen et al.). It is also expensive: long-form GST requires thousands of circuits and is practical up to about two qubits. For a quick average error rate, randomized benchmarking is far cheaper; for one targeted parameter, robust phase estimation suffices. Sandia’s open-source pyGSTi is the reference implementation, used across trapped-ion, superconducting, and spin-qubit platforms.

Key papers

Reference implementations

  • Quantum process tomography : The textbook protocol for fully characterizing a quantum gate: apply it to an informationally complete set of input states, tomograph the outputs, and reconstruct the complete channel.
  • Robust phase estimation : Heisenberg-limited calibration protocol that pins down individual gate rotation angles and axes from geometrically growing gate repetitions, with provable robustness to SPAM error.
  • Randomized Benchmarking : Estimates average gate error from the exponential decay of survival probability over increasingly long random Clifford sequences, robust to state-preparation and measurement errors.