Measure-Theoretic Quantum Measurement

Measurement theory distinguishes the probabilities of recorded outcomes from the transformations left on the measured system. POVMs describe the first; instruments describe both. This chapter supplies the measure-theoretic, operator-algebraic, and convergence assumptions needed for those statements, especially when the state space or outcome set is infinite.

The physical models and general working rules remain in Measurement and Open Quantum Systems. The articles here extend those treatments through precise theorems, proofs, and counterexamples rather than replacing their canonical introductions.

Enter this chapter

Start with density operators, normal states, and trace-class operators. A state is a positive trace-one operator, and bounded effects give its event probabilities. Normality ensures that countable event sums produce countably additive probability measures.

The chapter has three connected routes:

1. From events to observables. Effects and effect algebras develops operator order and partial addition. Observables as PVMs treats sharp outcome measures, while generalized observables as POVMs allows nonprojective effects. Naimark dilation proves that every POVM is a compressed PVM.
2. From probabilities to state changes. Completely positive maps explains matrix positivity and normal preduals. Kraus representations establishes convergence of countable operator sums. Instruments organizes operations into a measure, and state update maps analyzes conditioning and rare-event stability.
3. From a single apparatus to comparison and reconstruction. Joint measurability characterizes a common parent observable. Informational completeness asks whether statistics determine the input state. Unsharp measurements separates smearing and outcome variance from dynamical disturbance.

Readers familiar with finite matrices can enter the second route directly, but should check which topology is used whenever a finite sum becomes countably infinite.

The three mathematical objects

An effect is one operator $0\leq E\leq I$. Its expectation is the probability of one event.

A POVM assigns an effect $E(\Delta)$ to each measurable outcome set, with normalization and countable additivity. It determines all first-measurement statistics:

$$p_\rho(\Delta)=\operatorname{Tr}(\rho E(\Delta)).$$

An instrument assigns a CP operation $\mathcal I(\Delta)$ to each event. Its relation to the POVM is

$$E(\Delta)=\mathcal I(\Delta)^*(I), \qquad p_\rho(\Delta)=\operatorname{Tr}\mathcal I(\Delta)(\rho).$$

The normalized conditional state exists when this probability is positive. Otherwise the unnormalized output is zero and no posterior state is supplied by division.

These distinctions resolve several apparent paradoxes. The same POVM can have many instruments; an informative outcome need not prepare a pure state; and a continuous-outcome probability density does not require a normalized eigenvector at each point.

Dilation is a representation theorem

Naimark dilation writes $E(\Delta)=V^*P(\Delta)V$, with $V$ an isometry and $P$ a larger-space PVM. Stinespring dilation represents a normal CP Heisenberg map by $\Psi(A)=V^*(A\otimes I)V$. The first theorem concerns an outcome measure; the second concerns an operation.

Both explain why larger-space descriptions are useful. Neither chooses a unique apparatus Hamiltonian or a unique disturbance model from probabilities alone. Minimality supplies mathematical uniqueness up to an intertwining unitary, while unused auxiliary spaces permit many nonminimal realizations.

For countably many Kraus operators, completeness is generally a strong-operator limit. Output states converge in trace norm. Requiring operator-norm convergence of every completeness sum would rule out legitimate channels such as complete dephasing on an infinite basis.

A noisy-spin example links the routes

Consider

$$E_\pm=\tfrac12(I\pm\eta\sigma_z), \qquad0\leq\eta\leq1.$$

The effects form a binary POVM. It is sharp at $\eta=1$, a fair coin at $\eta=0$, and unsharp between those values. A valid instrument uses the square-root operators $\sqrt{E_\pm}$, but outcome-dependent output unitaries give different instruments with the same statistics.

The reported values $\pm1$ have first-moment operator $\eta\sigma_z$ and intrinsic-noise operator $(1-\eta^2)I$. Adding an equally noisy $\sigma_x$ measurement admits a joint parent precisely when $\eta\leq1/\sqrt2$. Nevertheless this single binary spin POVM is not informationally complete: it determines only one Bloch component.

Thus sharpness, compatibility, disturbance, and informational completeness answer different questions. None can be inferred from the others without additional hypotheses.

Checks before applying a formula

Identify the input and output Hilbert spaces, the outcome sigma-algebra, and whether the formula describes an event measure or a conditional density. For infinite systems, specify normality, countable additivity, and the relevant convergence topology. For postselection, retain the success probability. For unbounded reported values or energies, state the moment and domain assumptions.

The leaf articles supply proofs and solved exercises for these checks. Busch and collaborators provide the comprehensive measure-theoretic reference; Heinosaari and Ziman offer a complementary structural treatment, and Holevo connects the same language to infinite-dimensional quantum statistics and channels.

References

• Busch, Paul, Pekka Lahti, Juha-Pekka Pellonpää, and Kari Ylinen, Quantum Measurement, Springer, 2016 — general observables, operations, and instruments. Publisher.
• Heinosaari, Teiko, and Mário Ziman, The Mathematical Language of Quantum Theory: From Uncertainty to Entanglement, Cambridge University Press, 2012 — effects, compatibility, channels, and measurement models. Publisher.
• Holevo, Alexander S., Quantum Systems, Channels, Information: A Mathematical Introduction, 2nd revised and expanded ed., De Gruyter, 2019 — quantum statistics and infinite-system channels. Publisher.