Projection-Valued Measures

The spectral theorem hands you a Hermitian observable already broken into projectors. This lesson takes that decomposition seriously as the mathematical statement of measurement. A projection- valued measure (PVM) is the bookkeeping that assigns, to each possible outcome of an observable, a projector — and through the Born rule, a probability.

Projectors, recalled

A projector (orthogonal projector) is a Hermitian, idempotent operator:

P = P^\dagger, \qquad P^2 = P.

Idempotence says applying $P$ twice is the same as applying it once: projecting an already-projected vector changes nothing. The simplest example is the rank-one projector onto a unit vector $|k\rangle$ , namely $P = |k\rangle\langle k|$ , which sends $|\psi\rangle$ to the part of it along $|k\rangle$ : $P|\psi\rangle = \langle k|\psi\rangle\,|k\rangle$ . A projector's only eigenvalues are $0$ and $1$ — it answers a yes/no question about which subspace a state lies in.

Decomposing an observable

By the spectral theorem, a Hermitian observable $A$ with distinct eigenvalues $\{\lambda\}$ is

A = \sum_{\lambda} \lambda\, P_\lambda,

where $P_\lambda$ projects onto the eigenspace of $\lambda$ . The family $\{P_\lambda\}$ is the projection-valued measure of $A$ . It satisfies three defining properties:

P_\lambda = P_\lambda^\dagger = P_\lambda^2, \qquad P_\lambda P_{\lambda'} = 0 \ (\lambda \neq \lambda'), \qquad \sum_\lambda P_\lambda = \mathbb{1}.

Each $P_\lambda$ is a projector; different projectors are mutually orthogonal; and together they sum to the identity (completeness). This is the operator analog of a classical probability distribution partitioning the sample space — except the "events" are orthogonal subspaces rather than disjoint sets.

The measurement postulate, stated through projectors

Measuring the observable $A$ in the state $|\psi\rangle$ (assumed normalized) obeys:

Outcomes. The only possible results are the eigenvalues $\lambda$ .
Probabilities (Born rule). The probability of outcome $\lambda$ is

P(\lambda) = \langle\psi| P_\lambda |\psi\rangle = \| P_\lambda |\psi\rangle \|^2.

Collapse. Given outcome $\lambda$ , the post-measurement state is the renormalized projection

|\psi'\rangle = \frac{P_\lambda |\psi\rangle}{\sqrt{\langle\psi|P_\lambda|\psi\rangle}}.

Completeness $\sum_\lambda P_\lambda = \mathbb{1}$ guarantees the probabilities sum to one: $\sum_\lambda \langle\psi|P_\lambda|\psi\rangle = \langle\psi|\psi\rangle = 1$ .

Expectation values from the PVM

The expectation value of $A$ follows immediately by inserting the decomposition:

\langle A\rangle = \langle\psi|A|\psi\rangle = \sum_\lambda \lambda\,\langle\psi|P_\lambda|\psi\rangle = \sum_\lambda \lambda\, P(\lambda).

This is precisely the statistical mean of the outcomes $\lambda$ weighted by their probabilities — the bridge between the operator $A$ and the average of repeated measurements. The variance follows the same way from $\langle A^2\rangle = \sum_\lambda \lambda^2 P(\lambda)$ .

Example: measuring $Z$ on a qubit

The Pauli observable $Z = (+1)\,|0\rangle\langle 0| + (-1)\,|1\rangle\langle 1|$ has PVM $\{P_{+1} = |0\rangle\langle 0|,\ P_{-1} = |1\rangle\langle 1|\}$ . For $|\psi\rangle = \alpha|0\rangle + \beta|1\rangle$ , the outcome $+1$ has probability $\langle\psi|P_{+1}|\psi\rangle = |\alpha|^2$ and $-1$ has probability $|\beta|^2$ , with $\langle Z\rangle = |\alpha|^2 - |\beta|^2$ . This is the standard computational-basis measurement, now seen as the PVM of the observable $Z$ .

Why it matters

The projection-valued measure is the most general "sharp" (projective) measurement in quantum mechanics. It packages outcomes, probabilities, and collapse into one object derived directly from the spectral theorem. When you later meet generalized measurements (POVMs) and quantum channels, the PVM is the clean, projective special case they all reduce to — and the one every textbook measurement postulate is built on.