Noise models, T₁ and T₂, and noise characterization (randomized benchmarking, gate set tomography)
A quantum computer’s datasheet says “two-qubit gate fidelity 99.9%”. That one number hides a great deal. An error that randomly flips a qubit one time in a thousand, and an error that rotates every qubit by the same tiny angle every time, can have exactly the same average fidelity. After a thousand gates the first has caused about one error. The second has added up coherently into a large rotation. Error correction (Chapters 136–141) starts from an honest description of noise, and this chapter is about getting one.
On the plate: two kinds of noise with the same fidelity per gate, diverging as gates accumulate.
T₁ and T₂
Two times summarize a qubit’s memory. is the energy-relaxation time: an excited qubit decays to as (the amplitude-damping channel of Chapter 104). is the coherence time: a superposition’s phase is lost as . Decay itself destroys coherence, so , where is pure dephasing, and therefore . A Ramsey experiment (two pulses separated by a delay) measures , including slow drifts; adding a pulse in the middle, a Hahn echo, refocuses slow noise and measures . Set 100 µs and 200 µs. Established
Coherent versus stochastic
Over-rotate every gate by 10 milliradians. The infidelity per gate is . A stochastic error with the same per-gate probability accumulates linearly: after gates, infidelity . The coherent rotations add as amplitudes: . At and , stochastic noise gives 0.0025; coherent gives 0.23. Other failure modes escape the simple picture too: leakage out of the qubit subspace, and errors correlated across qubits or in time, such as those from cosmic-ray bursts in superconducting chips. Error-correction thresholds assume errors are local and not too correlated. Established
Randomized benchmarking and gate set tomography
To measure average gate quality without being fooled by state-preparation and measurement errors, Emanuel Knill and colleagues (2008) and Easwar Magesan, Jay Gambetta and Joseph Emerson (2011) developed randomized benchmarking (RB). Run random sequences of Clifford gates followed by the inverse, and fit the survival probability to
where is the average error per gate and . Random Cliffords twirl any noise into depolarizing noise, so RB reports one honest average, but it hides the coherent part that can be lethal in a structured algorithm. Gate set tomography (Robin Blume-Kohout and colleagues, 2017; Erik Nielsen and colleagues, 2021) reconstructs every gate, preparation and measurement self-consistently, revealing coherent errors at the cost of many more experiments. Established
What you now have
T₁, T₂ and T_φ and why , Ramsey and echo experiments, coherent versus stochastic accumulation, leakage and correlated noise, randomized benchmarking and its fitting model, and gate set tomography.
· · ·
Three questions to carry with you.
And one to work out. Each of 100 gates over-rotates by 0.01 rad. What is the infidelity ?
Sources: E. Knill et al., Phys. Rev. A 77, 012307 (2008). E. Magesan, J. M. Gambetta and J. Emerson, Phys. Rev. Lett. 106, 180504 (2011). R. Blume-Kohout et al., Nat. Commun. 8, 14485 (2017). E. Nielsen et al., Quantum 5, 557 (2021). M. McEwen et al., Nat. Phys. 18, 107 (2022) (cosmic-ray bursts). Checked by code: sin²(0.5) = 0.2298 vs 100 × 0.01²/4 = 0.0025.