Self-Assessment Quiz: Transpilation

Twenty questions. The routing and seed clusters are the ones that change practice. Aim for 16 or more.

Question 1

Which of these is not in a current IBM device's operation list? - A. ecr - B. rz - C. cx - D. sx

Question 2

A 127-qubit heavy-hex device has roughly how many qubit-to-qubit connections? - A. 127 - B. about 72 - C. about 8,000 - D. 16,000

Question 3

That sparsity is: - A. a manufacturing defect - B. deliberate — fewer couplings means less crosstalk and better coherence - C. a software limitation - D. temporary

Question 4

A SWAP gate costs how many CNOTs? - A. 1 - B. 2 - C. 3 - D. 6

Question 5

A 5-qubit all-to-all circuit with 10 logical CNOTs transpiled at optimization level 0 needed: - A. 10 two-qubit gates - B. 18 - C. 34 - D. 100

Question 6

The same circuit at level 2 needed: - A. 10 - B. 18 - C. 34 - D. 50

Question 7

A 5-qubit linear chain with 4 logical CNOTs transpiles to: - A. 4 two-qubit gates, at every optimization level - B. 12 - C. 18 - D. it varies unpredictably

Question 8

That means routing overhead is: - A. a constant tax on all circuits - B. proportional to how badly the circuit's connectivity mismatches the device's - C. always about 2× - D. only present at level 0

Question 9

The pass manager stage that decides which physical qubit each logical qubit maps to is: - A. init - B. layout - C. routing - D. translation

Question 10

FixedPoint and DoWhileController in the optimization stage mean the optimizer: - A. runs once - B. iterates until the circuit stops shrinking - C. is disabled - D. runs in parallel

Question 11

Optimization levels differ primarily in: - A. how many times they run the same passes - B. which algorithms they use - C. the random seed - D. nothing; they are aliases

Question 12

On a circuit whose connectivity already matches the device, higher optimization levels: - A. always reduce two-qubit gate count - B. leave two-qubit count unchanged and affect only depth - C. increase gate count - D. are required

Question 13

Can a higher optimization level produce a worse result? - A. never - B. yes — level 2 was measured producing a deeper circuit than level 1 - C. only at level 0 - D. only on simulators

Question 14

Which layout method assigns logical qubit $i$ to physical qubit $i$? - A. dense - B. sabre - C. trivial - D. VF2

Question 15

Compared with sabre routing, basic routing needed: - A. the same number of gates - B. about half as many - C. more than twice as many - D. it failed

Question 16

Routing to minimize SWAPs is: - A. solvable exactly in polynomial time - B. NP-hard, so all available methods are heuristics - C. handled by the hardware - D. unnecessary on modern devices

Question 17

A custom transpiler pass is a class implementing: - A. __call__(circuit) - B. run(dag) - C. transpile(qc) - D. optimize()

Question 18

Passes communicate through: - A. global variables - B. property_set - C. the circuit's metadata - D. they cannot communicate

Question 19

Transpiling one circuit at optimization level 3 with six different seeds produced two-qubit gate counts of: - A. all identical - B. 18 to 21 — a 17% spread - C. 10 to 100 - D. it does not depend on the seed

Question 20

Best-of-4-seeds on a routing-limited circuit bought: - A. nothing - B. about 10% fewer two-qubit gates for ~31 extra milliseconds - C. a 2× improvement - D. worse results

Answers

# Answer Why
1 C cx is not native; it is composed from ecr plus corrections. §10.1
2 B 144 directed edges, 72 connections — 0.90% of all-to-all. §10.1
3 B The routing overhead is the price of low gate error. §10.1
4 C No cheaper construction exists. §10.3, Ch. 4 §4.6
5 C Twenty-four extra gates: eight SWAPs. §10.3
6 B Nearly half, from compiler settings alone. §10.3
7 A Zero routing overhead — a chain maps onto a chain. §10.3
8 B And it widens with $n$ for a mismatched circuit. §10.3
9 B Routing then fixes what layout could not. §10.4
10 B Which is where higher levels spend their time. §10.4
11 B Not merely effort — different algorithms. §10.5
12 B Measured: 33 two-qubit gates at every level. §10.5
13 B Measured: depth 85 at level 2 vs 73 at level 1. §10.5
14 C Blind to both connectivity and error rates. §10.6
15 C 58 vs 28 — greedy, no lookahead. §10.6
16 B Which is why the method choice matters. §10.6
17 B Operating on the DAG, not the circuit. §10.8
18 B The pass manager's shared scratchpad. §10.8
19 B SABRE is randomized. §10.9
20 B The cheapest optimization in the book. §10.9

Topic Map

Questions Topic Section If you missed these
1–3 The target and topology §10.1 Do Exercise 10.6 on your own backend
4–8 Routing cost §10.3 Do Exercise 10.7. The chain-vs-all-to-all table is the chapter
9, 10 Pass manager stages §10.4 Reference; useful when a transpilation surprises you
11–13 Optimization levels §10.5 Level 2 as default; measure rather than assume
14–16 Layout and routing methods §10.6 Exercise 10.8; the 2× spread is worth seeing
17, 18 Custom passes §10.8 Exercise 10.10
19, 20 The seed §10.9 Do Exercise 10.11. Then fix your seeds forever

Score 16+: go to Chapter 11.

Score 12–15: the routing cluster (4–8) and the seed cluster (19–20) are the two that change what you do. If you missed either, run the corresponding exercise before continuing.

Score under 12: the pass and method details are reference you can look up. Three things to carry into Chapter 11: a SWAP costs three CNOTs, your gate budget is in transpiled gates, and always set seed_transpiler.