Passive measurement cannot choose every outcome
A number measurement can narrow a field into a particular number state, but the outcome need not equal the number requested in advance. Sayrin and colleagues added real-time feedback in 2011: atomic probes supplied information, a computer estimated the field, and a source applied small microwave displacements.
The new question is operational. When the observed record points away from the desired state, does an action change the subsequent distribution towards it? Merely relabelling a favourable subset of runs as successful prepares a conditional ensemble, with a success probability and waiting time that must also be reported.
Read the evidence: C. Sayrin et al. (2011), Real-Time Quantum Feedback Prepares and Stabilizes Photon Number States — author manuscript, arXiv:1107.4027; Nature 477, 73–77
The same average can hide very different states
Consider a target containing exactly two photons. A mixture giving equal probabilities to vacuum and four photons has the same mean number, yet it never contains two. A mean-value meter would declare no error while a number-resolving measurement would find no target population.
For a known pure target, its population in the density operator is the fidelity used here. This says how often the corresponding ideal projection succeeds. It is more informative than the mean for this task, but it does not by itself certify every off-diagonal element of an arbitrary unknown field state.
A unitary kick is not an exact photon repair
The 2011 actuator displaced the field in phase space; it did not inject exactly one photon whenever the estimator reported a loss. Success relied on alternating these controlled actions with number-sensitive measurements and updating the state estimate after both.
This distinction can be checked without simulating the apparatus. A displacement applied to vacuum produces a coherent state with a spread of photon numbers. If a proposed explanation says the same action always turns an arbitrary number state into the next one, it has replaced the actual actuator with a different operation.
Read the evidence: C. Sayrin et al. (2011), Real-Time Quantum Feedback Prepares and Stabilizes Photon Number States — author manuscript, arXiv:1107.4027; Nature 477, 73–77
The estimator is not a second detector
A filtered state uses more than the latest atomic bit. Preparation, probe calibration, applied actions, relaxation and detection imperfections enter its construction. An undetected atom can still have interacted with the field. Flight time also means that some probes have acted before their outcomes become available.
For a general audit, plot three streams separately: the raw readout, the estimated distribution and the commanded action. A delayed correction can be reasonable even when the current estimate already looks improved. Without timestamps, it is easy to misread a causal control sequence as an instantaneous reaction to a fully known state.
Read the evidence: C. Sayrin et al. (2011), Real-Time Quantum Feedback Prepares and Stabilizes Photon Number States — author manuscript, arXiv:1107.4027; Nature 477, 73–77
Verification must say when a run stops
The paper compared ensembles stopped after a fixed time with ensembles stopped after an estimated target threshold was sustained. Additional probe atoms after interruption supplied a photon-number reconstruction independent of the feedback estimator, while still relying on calibrated probing. The two stopping rules answer different performance questions.
A threshold-selected ensemble can have higher final fidelity because the procedure waits for favourable evidence. A fixed-time ensemble includes trajectories still recovering from a jump. Comparing their bars as if they were the same protocol hides the cost in time and the definition of success.
Read the evidence: C. Sayrin et al. (2011), Real-Time Quantum Feedback Prepares and Stabilizes Photon Number States — author manuscript, arXiv:1107.4027; Nature 477, 73–77
Stabilization means recovery, not immortality
In an ideal vacuum-loss channel, each photon has the same independent survival probability. A state with a fixed number therefore has a first-loss rate proportional to that number. Larger target numbers give a controller less time to acquire information and act; thermal gain and experimental errors add further demands.
The linked qubit models separate damping, state reconstruction and error correction. They do not implement this oscillator feedback loop. Restoring a known number-state target also differs from protecting arbitrary unknown encoded information: learning the number is acceptable for the first task and may destroy information required by the second.