What does quantum nondemolition readout preserve, and what does it necessarily disturb?

Source research checked 17 Oct 2026. Teaching calculations are ideal models.

A dispersive measurement can leave a qubit in its original energy eigenstate while still removing the coherence of a superposition. This distinction is central to interpreting the phrase quantum nondemolition. The protected quantity is a specified observable under a specified interaction. In an ideal energy-basis measurement, repeated measurements can agree because the measurement preserves the populations of the energy eigenstates. Agreement does not imply that the apparatus preserved every property of an arbitrary input. A device that learned which energy state was present generally also changed the relative-phase information that made an initial superposition useful.

Consider the joint state of the qubit and a readout resonator before a classical label is assigned. An initial qubit superposition can evolve into a superposition of two different resonator responses, one associated with each qubit state. The resonator has become a pointer. If those responses are identical, they carry no information about the qubit. If they are distinguishable, a measurement of the outgoing field can reveal the qubit state. Looking only at the qubit requires averaging over the unobserved field. Its remaining coherence is then multiplied by the overlap between the two possible pointer states.

This picture separates the physical measurement from the classifier displayed on a screen. Throwing away the digital result does not reunite the two outgoing microwave fields. Moving the decision threshold also does not reverse information that has already escaped. A deliberately weak measurement allows the pointer responses to overlap substantially, leaving a partial measurement record and partial coherence. A strong measurement separates the responses more clearly. In a realistic apparatus, the detector may collect only some of the available information, so the qubit can suffer backaction that the recorded data do not fully explain.

Repeated-readout tests therefore need several observables. Prepare an energy eigenstate, measure it, and ask whether a later measurement reports a transition. Separately, prepare a superposition and measure its interference contrast after a controlled readout pulse. These experiments diagnose population changes and dephasing through different preparations. A high assignment accuracy can coexist with appreciable state transitions; a low transition probability can coexist with complete dephasing. Neither one-number summary replaces the other. The duration between the two measurements must also be reported because ordinary relaxation during the wait can imitate demolition by the first pulse.

Residual resonator photons make timing part of this story. Even after the main readout drive stops, a field left in the resonator can continue interacting with the qubit. Subsequent gates may then accumulate photon-dependent phase shifts. A reset or depletion procedure must be evaluated together with the measurement sequence, not inferred from an empty-looking waveform trace. The useful question is whether the field seen by the qubit has returned sufficiently close to the intended state before computation resumes. This matters especially when a measured ancilla is reused while nearby data qubits must remain coherent.

Collection efficiency provides a final consistency test. Imagine leaving the qubit–field interaction unchanged while discarding part of the outgoing signal before detection. The accessible record becomes less informative, but the qubit has already become correlated with the full field. Its dephasing need not decrease with the amount of information retained by the observer. This is why a poor detector can cause substantial measurement backaction without delivering a correspondingly useful result. Conversely, an efficient record can support better conditional state estimation because more of the disturbance is accounted for. These comparisons require the same interaction and preparation. Changing the pulse while changing the detector would mix information collection with a different physical measurement strength and obscure the distinction.

Follow the mechanism

1. Prepare a qubit superposition and a common initial resonator state, so any later pointer separation is produced by the interaction rather than different initial conditions.
2. Condition the resonator response on the energy-basis qubit value. The resulting qubit-field state contains correlations before a digitizer makes a decision.
3. Trace out the field to obtain the qubit state alone. The off-diagonal term acquires the pointer-state overlap, while ideal population-preserving coupling leaves diagonal terms unchanged.
4. Compare an energy-repeatability experiment with a Ramsey-style coherence experiment. Add detector inefficiency only to the accessible record, not as an automatic removal of backaction.

Before interpreting a result

• A separately measured assignment matrix and induced-transition matrix.
• A coherence experiment with the readout pulse applied to the target or a spectator.
• Timing information for ring-down, depletion and reuse.

A measurement can preserve the observable it reads while consuming the coherence of a superposition. Readout evaluation becomes clearer when assignment, transitions and information loss are reported separately.