Generalized Measurements and Instruments

A realistic detector can be noisy, inefficient, indirect, or destructive. Generalized measurement theory describes its outcome statistics with a POVM and its conditional state changes with an instrument. The distinction matters whenever the measured system is used again: two detectors can report the same probabilities and leave different states.

Required background. Measurement Theory, Measurement as an Operation, and State Update Rules introduce outcomes and conditioning. Use Density Operators for mixed inputs and outputs.

Probabilities and backaction are separate data

For a discrete POVM, $E_r\ge0$, $\sum_rE_r=I$, and $p_r=\operatorname{Tr}(\rho E_r)$. A compatible instrument has completely positive operations satisfying

$$\begin{gathered} \mathcal I_r(\rho)=\sum_\alpha K_{r\alpha}\rho K_{r\alpha}^\dagger,\\ E_r=\sum_\alpha K_{r\alpha}^\dagger K_{r\alpha}. \end{gathered}$$

The index $r$ is reported; $\alpha$ is unresolved. The trace of $\mathcal I_r(\rho)$ is $p_r$, and its normalized output is defined only for $p_r>0$. Averaging over all $r$ gives a trace-preserving channel.

The positive-square-root choice $K_r=\sqrt{E_r}$ is one possible instrument. Replacing it by $U_r\sqrt{E_r}$ preserves the effects while changing the output through an outcome-dependent unitary $U_r$. More general instruments can reset the state or change the output Hilbert space. Effects alone therefore cannot answer a backaction question.

From detector statistics to a physical realization

Use One Interaction, Different Measurement Records to compare two detector bases after the same ancilla interaction. Their conditional states and outcome statistics differ, while their averaged channel agrees.

Start with Why Generalized Measurements Are Needed, then POVMs, Kraus Operators, and Quantum Instruments. These four pages supply the main calculation tools.

Two dilation theorems explain compatibility with a larger quantum system:

Question	Construction
Can a POVM be obtained from a sharp measurement on a larger space?	Naimark Dilation gives $E_r=V^\dagger\Pi_rV$ for an isometry $V$ and projectors $\Pi_r$.
Can a channel be obtained by discarding an initialized environment after joint evolution?	Stinespring Dilation gives $\Phi(\rho)=\operatorname{Tr}_E(V\rho V^\dagger)$ for an isometry $V$.

The isometry form in the second row assumes trace preservation. A general trace-decreasing operation needs the corresponding more general dilation or a selected outcome of a larger channel. Neither theorem identifies a unique laboratory apparatus from the effective object alone.

Unsharp statistics do not guarantee gentle dynamics

The effects

$$E_\pm=\frac12(I\pm\lambda Z), \qquad 0\le\lambda\le1,$$

interpolate from an uninformative random outcome to a sharp $Z$ measurement. The parameter $\lambda$ describes the statistics, not the full disturbance. A strong unitary kick or a destructive reset can accompany the same effects even when $\lambda$ is small.

Unsharp Measurements studies such effects and their compatible instruments. Weak Measurements adds a controlled small-coupling regime. Weak Values examines the first-order pointer response after postselection; a large weak value is neither an ordinary eigenvalue nor a guarantee that the first-order approximation remains valid. Protective Measurements adds protection and adiabatic assumptions that weakness alone does not supply.

What a detector calibration can determine

In Measurement Tomography, known probe states $\rho_j$ and outcome frequencies constrain $\operatorname{Tr}(\rho_jE_r)$. Informationally complete probes can determine the effects in an ideal noiseless model. Finite data require uncertainties and physical constraints such as positivity and completeness.

Those outcome frequencies do not determine the conditional output maps. Instrument tomography also probes the output states. If the eventual experiment involves sequential measurements or feedback, calibrating only the POVM can leave the needed backaction unspecified.

Keep the no-click or failure outcome whenever it is part of the experiment. Renormalizing successful trials describes a selected ensemble and changes the question. For time-resolved records, continue to Continuous Measurement and Quantum Trajectories; for ignored records, continue to Quantum Channels and Noise.

Exercises

A reset hidden behind sharp statistics. Compare $K_r=|r\rangle\langle r|$ with $R_r=|+\rangle\langle r|$ for $r=0,1$. Find their effects and the probability of obtaining $0$ in a later $Z$ measurement, conditioned on first obtaining $r=0$.

Solution

Both have effect $|r\rangle\langle r|$, and both families are complete. When the first outcome has nonzero probability, the first implementation leaves $|0\rangle$ and the second leaves $|+\rangle$. The later probability is respectively one and $1/2$. Identical first-outcome statistics do not imply identical sequential statistics.

A forgotten fine outcome. If a device reports only $r$ after physically distinguishing $(r,\alpha)$, how is its operation obtained?

Solution

Sum the fine operations: $\mathcal I_r=\sum_\alpha\mathcal I_{r\alpha}$. Their effects likewise add. One sums completely positive maps, not probability amplitudes for already distinguished records. The resulting conditional state can be mixed because $\alpha$ is unavailable, without any additional interaction caused by forgetting it.

References

• Busch, P., P. Lahti, J.-P. Pellonpää, and K. Ylinen. Quantum Measurement. Springer, 2016. Generalized measurements, instruments, and realizations.
• Kraus, K. States, Effects, and Operations: Fundamental Notions of Quantum Theory. Springer, 1983. Completely positive operations and measurement models.
• Watrous, J. The Theory of Quantum Information. Cambridge University Press, 2018. Measurements, channels, and dilation representations. Author manuscript.
• Wiseman, H. M., and G. J. Milburn. Quantum Measurement and Control. Cambridge University Press, 2010. Weak measurements and conditional dynamics.