Quiz: Pulse-Level Programming, and What Replaced It

Answers with explanations at the end.

1. What happened to qiskit.pulse, and when?

2. Name three other things removed alongside it.

3. Give the three reasons the pulse abstraction was removed.

4. What is target.dt, and what does it constrain?

5. Describe what a gate physically is on a superconducting device, and say what sets the rotation angle, the axis, and the leakage.

6. rz has a duration of 0.0 ns. Explain.

7. Why does the virtual Z gate matter for reading Chapter 28's transpiled circuits?

8. Which is the slowest operation on this device, and by how much?

9. How many two-qubit gates fit inside $T_1$, and what fraction of them survive at the measured gate error?

10. Which binds first on this device — gate error or decoherence? Is that universal?

11. A scheduled 6-qubit ansatz runs in 7.10 μs against a $T_2$ of 170 μs. What fraction of qubit-time is idle, and why is that number so high?

12. What is the difference between ASAP and ALAP scheduling, and which is preferable for a circuit ending in measurement?

13. What does dynamical decoupling do, and what is the physical mechanism?

14. Give the measured effect of XX and XY4 decoupling on the ansatz's fidelity.

15. Those measurements were significant at 2σ across six seeds. Why is the obvious conclusion wrong?

16. Precisely why can dynamical decoupling not help against Markovian noise?

17. Every quality marker of a good measurement was present in §31.4. What does that tell you about the limits of statistical rigour?

18. State Chapter 26's rule about simulators and hardware, and the corollary this chapter adds.

19. Give the six instances from Part V of the same failure pattern.

20. Order Part V's four levers by what each is worth, and say what is notable about the order in which they are usually attempted.

Answers

1. It was deprecated across the Qiskit 1.x series and removed in 2.0. In Qiskit 2.5.1 import qiskit.pulse raises ModuleNotFoundError.

2. Any three of: QuantumCircuit.calibrations, QuantumCircuit.add_calibration, backend.defaults, backend.instruction_schedule_map, backend.drive_channel.

3. (a) The abstraction did not port — a pulse schedule is meaningless on another backend and absurd on trapped ions, while Qiskit's other abstractions survive a hardware change. (b) The user base was tiny and largely internal to vendors, who had better device-specific tools. (c) The hardware moved past it — fractional gates, parametrized native operations, and richer Target metadata now cover most cases.

4. The sample period of the waveform generator driving the qubits — 0.2222 ns here. Every duration on the device is an integer multiple of it, so it is the finest time resolution anything, including the scheduler, can express.

5. A shaped microwave pulse applied near the qubit's transition frequency for a precisely controlled duration. Area sets the rotation angle; phase sets the rotation axis; envelope shape determines leakage into $|2\rangle$, which is why real pulses use DRAG corrections rather than square envelopes.

6. It is the virtual Z gate. A $Z$ rotation is not performed — it is absorbed into the phase of every subsequent pulse, with the control software redefining what "the $X$ axis" means from that point on. Exact, instantaneous, and error-free.

7. Because transpiled circuits are full of rz and it costs nothing. Chapter 29's ansatz has 83 rz gates contributing zero time and zero error. The basis $\{$rz, sx, x, ecr$\}$ is one free gate, two fixed 56.9 ns pulses, and one expensive entangler.

8. measure, at 1,216 ns — 2.3× the median two-qubit gate (533.3 ns) and over 20× a single-qubit gate. This is one reason Chapter 25 §25.7's syndrome-extraction cycle is hard: it is measurement-bound, not gate-bound.

9. $T_1 / t_{\text{ecr}} = 278.4\ \mu\text{s} / 533.3\ \text{ns} \approx$ 522 gates. At Chapter 12's measured error of 0.0075, $(1-0.0075)^{522} \approx$ 0.020 — about 2%.

10. Gate error, decisively — you lose the signal to imperfect gates long before running out of coherence time. Not universal: early superconducting devices were decoherence-limited, and trapped ions have very long coherence with slow gates. Check which regime you are in before optimizing for the wrong one.

11. 97% of qubit-time is idle (873.40 μs of 901.30 μs), and the circuit uses only 4.17% of the $T_2$ budget. The fraction is high because at any moment most of the circuit's qubits are waiting for a gate elsewhere — enormous coherence headroom, almost all of it spent doing nothing.

12. ASAP starts each instruction as soon as its inputs are ready, accumulating idle time at the end of a qubit's life; ALAP delays everything as long as possible, accumulating it at the beginning. For a circuit ending in measurement, ALAP generally leaves less idle time immediately before readout, which is when it hurts most.

13. It fills idle windows with pulse sequences that are the identity overall — XX, or the four-pulse XY4. The mechanism is refocusing: flip the qubit halfway through the window so the phase accumulated in the second half cancels the phase accumulated in the first.

14. XX: $-0.0053 \pm 0.0012$. XY4: $-0.0057 \pm 0.0012$. Both significantly worse, and a deliberate 20 μs idle window widened the gap.

15. Because the result is a property of the simulator, not of the technique. AerSimulator.from_backend builds a purely Markovian noise model, and DD has nothing to work with there — so it can only add the error of its own $X$ gates, which it does, reproducibly and meaninglessly.

16. Refocusing requires the phase-accumulation rate to be roughly the same in both halves of the idle window — that is, the noise must be correlated over the timescale of the sequence. Real devices have exactly that ($1/f$ flux noise, slow frequency drift, static spectator coupling). A memoryless channel's second half is statistically independent of its first, so flipping the qubit cancels nothing while the flip itself carries error.

17. That statistical rigour cannot rescue a model that lacks the phenomenon under study. Six seeds, a standard error, multiple idle lengths, 2σ testing, and two sequences agreeing made the wrong answer more convincing, not less. A well-executed measurement of the wrong model is a confident wrong answer, and it comes with error bars.

18. Chapter 26 §26.1: "Debug on a simulator, where you can see everything; validate on hardware, where you can see nothing." The corollary: some questions cannot be asked of a simulator at all.

19. Ch. 25 — a QEC test storing an eigenstate of its own failure mode. Ch. 26 — a bisection whose default input could not see the bug. Ch. 27 — three of four oracle-free properties passing a broken circuit. Ch. 29 — a layout chosen by connectivity alone, routing through dead edges. Ch. 30 — a benchmark robust to readout error, silent when readout error broke everything. Ch. 31 — a simulator with no correlated noise, judging correlated-noise cancellation. In every case the blindness was a documented, deliberate property of the method.

20. Shape (Ch. 29, +0.1658) > optimization (Ch. 28, +0.0262) > scheduling (Ch. 31) > mitigation (Ch. 13, −79% of residual error). Each layer is worth roughly an order of magnitude less than the one above. What is notable is that they are attempted in practice in the opposite order — because mitigation is a library call and shape is a design decision.