Skip to content
Tech Interview Prep
Technical interview guide

Quantum Error Correction, Noise & Decoherence

Why real qubits are noisy, the difference between physical and logical qubits, and how surface codes trade physical qubit count for error resilience.

Read
55 min
Practice MCQs
25
Interview Q&A
25
Edition
v2
Editorial status
Reviewed

Scope: IBM Quantum current QEC, noise, mitigation, and fault-tolerance guidance; surface-code and QEC literature reviewed 2026-09-04.

Curated: · Written: · Reviewed:

25 questions

MCQ-1A hypothetical 10-qubit superconducting processor prepares a GHZ state using 9 CZ gates and 10 single-qubit gates. Calibration reports 0.8% error per CZ, 0.1% error per single-qubit gate, and 2% readout error per qubit. Assuming independent errors and no other noise, what is the expected GHZ state fidelity from the gate errors alone, before readout?
MCQ-2A transmon is excited to |1> with a π pulse, held idle for a swept delay, then read out. The measured P(|1>) falls as exp(−t/47 µs), and the same 47 µs is fitted whether the idle window sits before or after a two-qubit gate. What does the 47 µs quantify?
MCQ-3A superconducting qubit measures T1 = 80 µs by inversion recovery, T2* = 12 µs by Ramsey, and T2_echo = 70 µs by a Hahn echo. What does the gap between the Ramsey and echo results establish about the Ramsey decay?
MCQ-5A team models a superconducting device with T1 = 60 us and T2 = 45 us. Their simulator applies a depolarizing channel with p = 0.001 to every single-qubit gate (0.1% infidelity, matching what they measure on hardware) and applies no error to qubits that are not gated. They run a circuit in which qubit Q5 is prepared in |1> and then parked, undriven, for 20 us while gates run on the other seven qubits. The simulator gives a state fidelity of 1.00 for Q5; the measurement on the device returns 0.72. Which explanation of the gap is correct?
MCQ-4A 200-gate circuit starts in |0> and uses one-qubit gates that each carry a coherent over-rotation of 0.1° about the X axis. Randomized benchmarking on the same gates reports an average error per gate of 5×10^-7. The team models the noise as depolarizing and predicts a final error probability of 200 × 5×10^-7 = 1×10^-4. What final error should they actually expect?
MCQ-25During fault-tolerant quantum error correction, syndrome extraction circuits take a finite cycle time \(t_\text{cycle}\) consisting of multi-qubit entangling gates, ancilla readout, and state reset. How does \(t_\text{cycle}\) constrain the system relative to physical decoherence times \(T_1\) and \(T_2\)?
MCQ-22A transmon idles for 20 µs waiting on its gate partner. During that window T2 falls to 8 µs while T1 holds at 60 µs, and the dominant noise is low-frequency 1/f flux dephasing. The team inserts an XY4 pulse sequence into the idle window. What does this primarily accomplish?
MCQ-23A team runs a 20-qubit device whose two-qubit gate error is 0.8% per gate. They want the expectation value of a 12-gate-deep observable circuit. Running it directly gives a value biased by roughly 0.8% × 12 ≈ 9.6%. They run the same circuit at noise scale factors 1, 2, and 3 (by inserting pairs of identity gates), fit the three measured expectation values to a linear function of the scale factor, and report the intercept. They do not add any ancilla qubits, do not measure any stabilizers, and do not decode any syndrome. Which statement best describes what this procedure does and why it is not quantum error correction?
MCQ-11Two logical states go through the same three-qubit bit-flip encoder: |0_L> = |000> and |+_L> = (|000> + |111>)/sqrt(2). Each run then gets one noiseless syndrome round: a fresh ancilla measures Z1Z2, a second ancilla measures Z2Z3, and the two ancilla bits are read out (0 for +1, 1 for -1). The team takes 1,000 shots per state and compares the two readout records to argue about no-cloning. Which reading of the records is correct?
MCQ-8A 5-qubit repetition code encodes one logical bit in physical qubits 1–5 and measures the four stabilizer generators Z1Z2, Z2Z3, Z3Z4, Z4Z5 each round. One round returns (Z1Z2, Z2Z3, Z3Z4, Z4Z5) = (+1, −1, −1, +1). The decoder uses minimum-weight decoding. What does it conclude and apply?
MCQ-12A team is building a distance-3 quantum memory. Their first design uses a single classical [7,1,3] Hamming code, applying the same parity checks to detect both bit flips and phase flips. In simulation, bit-flip errors are corrected, but a single Z error on any qubit goes undetected and corrupts the logical state. They want to keep the same 7 physical qubits and the same classical code, but detect and correct both error types. Which change to the check structure fixes the problem?
MCQ-7A team proposes a 4-qubit stabilizer code with generators g1 = X X X X, g2 = Z Z Z Z, g3 = X Z X Z, g4 = Z X Z X. They claim the four generators are independent and define a code. Which statement about this set is correct?
MCQ-9A stabilizer code's logical operators have been enumerated. The minimum-weight logical X operator acts on 3 qubits, the minimum-weight logical Z operator acts on 4 qubits, and no nontrivial logical operator has lower weight than these. What is the code distance, and how many arbitrary single-qubit Pauli errors can this code correct?
MCQ-6During a surface-code experiment, each data qubit is subjected to a coherent rotation R_x(θ) with θ = 0.05 rad per cycle. The stabilizer measurements are perfect. The team notices that the syndrome is trivial in most cycles, but occasionally it flags a single X error. What does this pattern indicate?
MCQ-17A distance-3 surface code measures a weight-4 Z-type stabilizer with one ancilla prepared in |0> and four CNOTs (ancilla as control, data as target) applied in the order d1, d2, d3, d4. During one extraction round the ancilla picks up a single X error after the CNOT on d1 and before the CNOT on d2. What does this round leave in the data qubits?
MCQ-14A distance-5 surface code is decoded with minimum-weight perfect matching. Every data qubit is error-free, but in round 3 one ancilla measurement on stabilizer S returns the wrong bit. The decoder is first given the syndrome changes for rounds 2–3 only, then re-run with round 4 included. What does it do in each case?
MCQ-15A team runs a surface-code memory on hardware with 0.1% two-qubit gate error and 1% readout error. Their decoder builds a 2D minimum-weight perfect matching graph independently from each round's syndrome. Measured logical error rate per round: 2.1e-3 at distance 3 and 2.0e-3 at distance 5. What explains the flat curve?
MCQ-13A team runs memory experiments on a superconducting grid where every stabilizer measurement couples nearest-neighbour qubits, so each rotated surface-code patch is laid out in two dimensions with no long-range gates. With their current gate and readout errors they measure a logical error rate per syndrome-extraction cycle of 1.60% at distance 3 and 0.75% at distance 5, using the same decoder in both runs. Keeping the noise and the decoder fixed, what per-cycle logical error rate should a distance-7 patch show?
MCQ-10A team's surface-code experiment uses physical two-qubit gates with a 1.5% error rate and a standard minimum-weight perfect-matching decoder. Measured logical error rate per cycle: 4×10⁻³ at distance 3, 1.2×10⁻² at distance 5, 3×10⁻² at distance 7. What does this trend indicate?
MCQ-20A team runs a distance-5 surface code with minimum-weight matching. At a physical error rate of 0.3% per gate they measure a logical error rate of 1.2×10⁻³ per round, and distance sweeps fit a threshold of 0.6% for this code, decoder and noise model. They then improve their two-qubit gates so the physical error rate drops to 0.1% per gate, changing nothing else. Sub-threshold scaling for this setup is ε_L ≈ A(p/p_th)^{⌊(d+1)/2⌋}, with A fixed by the code, decoder and noise model. What logical error rate per round should they expect?
MCQ-21A team reports a distance-5 surface-code logical qubit built from 49 physical qubits with a logical error rate of 1.0×10⁻³ per cycle, and their best single physical qubit measures 1.2×10⁻³ per cycle. They call this break-even. Which comparison would make the claim defensible?
MCQ-18A logical CNOT between two distance-3 code blocks is implemented transversally: physical CNOTs are applied between corresponding qubits of the two blocks. A single physical X error occurs on qubit 3 of the control block immediately before the gate. What error configuration do the two blocks carry afterward?
MCQ-16A d=5 surface-code logical qubit runs 25 syndrome-extraction rounds, then a logical T gate implemented by magic-state injection, then 25 more rounds, then a logical Z measurement. The team tracks the Pauli frame classically and applies no physical Pauli corrections until the final measurement. The measured logical error rate is ~0.5 per run, while the decoder predicts ~1e-3 for the Clifford-only segments. What is the cause?
MCQ-19A surface-code architecture compiles every logical circuit into Clifford gates plus T gates, and its T gates are produced by injecting distilled magic states from a factory. The team measures the factory's output error at 1e-3 per magic state, while the target logical error per T gate is 1e-6. They propose to fix this by widening the surface-code distance of the data qubits from 15 to 25. What is the most likely outcome of that change?
MCQ-24A team budgets a surface-code machine for a chemistry simulation: 100 logical qubits, 10^9 T gates, code distance d = 21 (one logical qubit occupies 2d² = 882 physical qubits), and 4 T-state factories, each occupying 12 logical patches and emitting one T state every 10 surface-code cycles. One cycle is 1 µs. Their draft lists 88,200 physical qubits and a runtime of 10^9 cycles ≈ 17 min. Which budget is right?