Quantum Measurement

Measuring a qubit forces it to pick a side: projective measurement collapses a superposition of |0⟩ and |1⟩ into one definite classical outcome, and that collapse is irreversible — there is no way to recover the original superposition from the classical bit that comes out the other end. This is not a limitation of today's hardware, it is a structural feature of quantum mechanics itself, and it is the reason a quantum computer cannot simply be read out mid-circuit the way a classical register can; any measurement destroys the quantum information the algorithm was carrying.

In practice the physical act of measuring a qubit is itself a noisy process, and on today's NISQ-era superconducting and trapped-ion hardware that readout error is one of the largest error budgets in the whole system. A superconducting qubit is typically read out by coupling it to a microwave resonator and inferring the qubit state from the resonator's dispersive shift, and that inference is a statistical discrimination problem — the measured signal for |0⟩ and |1⟩ overlaps somewhat, so a single shot is sometimes assigned to the wrong bit. IBM's and Google's superconducting processors report single-qubit readout fidelities typically in the 98-99.5% range, and even Google's 2023-era Sycamore-class devices and IBM's Heron processors, despite steady improvement, still show measurement error as one of the dominant noise sources alongside two-qubit gate error — a readout fidelity of 99% sounds high, but on a 100-qubit circuit measured every shot, uncorrected readout error alone compounds enough to meaningfully distort the output distribution.

Because a single measurement is a noisy coin flip rather than a clean readout, quantum algorithms run every circuit many times — typically thousands of shots — and build a probability histogram instead of trusting any one result. This repeated-shot requirement is not just about quantum uncertainty in the underlying state; it is also how experimenters average out readout error statistically, and labs further apply measurement-error mitigation by first characterizing the device's confusion matrix (how often a true |0⟩ reads out as |1⟩ and vice versa) on calibration circuits, then inverting that matrix to correct the raw histogram. Error-correcting codes such as the surface code push this further by measuring ancilla qubits repeatedly to detect errors without collapsing the encoded logical qubit itself, but those syndrome measurements are themselves imperfect and have to be repeated across several rounds before a decoder trusts the result — readout noise does not go away under error correction, it just becomes one more error channel the decoder has to model.

The irreversibility of collapse is what makes measurement timing the most consequential decision in a quantum circuit's design, not an afterthought bolted on at the end. Mid-circuit measurement, used in dynamic circuits and in every real-time error-correction cycle, has to budget for the fact that the act of measuring one qubit can disturb neighboring qubits through residual coupling, and the measurement itself takes hundreds of nanoseconds to microseconds on superconducting hardware — long enough that other qubits are idling and accumulating decoherence while they wait. Rigetti, IBM, and Google each publish readout duration and fidelity as headline specs precisely because the number trades directly against both circuit depth and total run time, and any claim of quantum advantage has to account for how much of the final answer's noise came from the algorithm versus from simply reading the answer out.