Observable

An observable specifies a set of possible measurement outcomes and an operator-valued rule that assigns their probabilities in every quantum state. A sharp real-valued observable is represented by a self-adjoint operator, equivalently by its projection-valued spectral measure. A generalized observable is represented by a positive-operator-valued measure, or POVM.

The statement “observables are Hermitian matrices” is correct for finite-dimensional sharp observables, but it is not the full definition.

In Symbols

Discrete Sharp Observable

For a sharp observable $A$ with discrete spectrum,

A=\sum_a a\Pi_a,

where the real outcome values $a$ and orthogonal spectral projectors $\Pi_a$ satisfy

\Pi_a^\dagger=\Pi_a, \qquad \Pi_a\Pi_b=\delta_{ab}\Pi_a, \qquad \sum_a\Pi_a=I.

In state $\rho$ , the Born probability of outcome $a$ is

p(a)=\operatorname{Tr}(\rho\Pi_a).

The expectation value and variance are

\langle A\rangle_\rho =\operatorname{Tr}(\rho A) =\sum_a a\,p(a),

(\Delta A)^2_\rho =\operatorname{Tr}(\rho A^2) -\langle A\rangle_\rho^2.

An expectation value is a distributional average. It need not be one of the possible outcomes.

General Sharp Observable

The spectral theorem represents a possibly continuous self-adjoint observable as

A=\int_{\mathbb R}a\,P_A(da),

where $P_A(\Delta)$ is the projector associated with a measurable outcome set $\Delta\subseteq\mathbb R$ . The probability measure in state $\rho$ is

\Pr_\rho(A\in\Delta) =\operatorname{Tr}\!\left[ \rho P_A(\Delta) \right].

This formulation covers discrete, continuous, and mixed spectra. Generalized eigenvectors such as $\lvert x\rangle$ are useful representations of continuous spectra, but they are not ordinary normalizable Hilbert-space eigenvectors.

Generalized Observable

A POVM $E$ assigns a positive operator $E(\Delta)$ to each measurable outcome set, with

E(\Delta)\ge0, \qquad E(\Omega)=I,

and countable additivity on disjoint sets. Its probability rule is

\Pr_\rho(\Delta) =\operatorname{Tr}\!\left[ \rho E(\Delta) \right].

For finitely many outcomes $k$ ,

E_k\ge0, \qquad \sum_kE_k=I, \qquad p(k)=\operatorname{Tr}(\rho E_k).

The effects $E_k$ need not be projectors or mutually orthogonal. A PVM is the sharp special case.

Canonical Home

See Observables for the canonical conceptual treatment, including apparatuses, instruments, coarse graining, and generalized measurements.

Eigenvalues and Eigenstates treats sharp values, Spectra distinguishes spectral types, and Spectral Decomposition develops the operator representation.

Roles That Must Be Separated

Object	Role
Observable	Specifies outcome labels and their probability operators
Self-adjoint operator $A$	Represents a sharp real-valued observable
PVM $P_A$	Gives the full sharp outcome-event structure
POVM $E$	Represents a possibly unsharp or generalized observable
Apparatus	Physical device implementing a measurement procedure
Instrument	Gives both outcome probabilities and conditional state changes
Outcome $a$	Recorded value or label in one trial
Expectation $\langle A\rangle$	Mean of the outcome distribution over repeated trials

An observable is not the same thing as a laboratory device. Different devices can implement operationally equivalent observables, and one device can be reconfigured to implement different observables.

An observable also does not by itself determine state update. A PVM or POVM fixes probabilities. A quantum instrument specifies the associated conditional transformations.

Self-Adjointness and Domains

In finite dimensions, a Hermitian matrix satisfies

A^\dagger=A

and is automatically a bounded self-adjoint operator. In infinite dimensions, an unbounded operator has a domain $D(A)$ , and the condition

A=A^\dagger

includes equality of both actions and domains. A merely symmetric operator can fail to be self-adjoint and may not generate the spectral measure required for a sharp observable.

Boundary conditions can be part of the observable. The same differential expression for momentum on different configuration spaces or domains can define different operators with different spectra.

Hermitian versus Self-Adjoint and Unbounded Operators give the canonical domain cautions.

Degeneracy and Sharp Values

If several linearly independent vectors share eigenvalue $a$ , the outcome is represented by the projector onto the whole eigenspace:

\Pi_a =\sum_{\lambda} \lvert a,\lambda\rangle \langle a,\lambda\rvert.

A state has a sharp value $a$ when it lies entirely in that spectral subspace:

\Pi_a\rho\Pi_a=\rho.

For a pure state, this reduces to $A\lvert\psi\rangle=a\lvert\psi\rangle$ . A nondegenerate eigenvector is only the simplest case.

Functions, Calibration, and Units

A real function $f$ of a sharp observable is defined spectrally:

f(A)=\int f(a)\,P_A(da).

This represents a deterministic relabeling of each outcome $a$ by $f(a)$ . If $f$ is not one-to-one, distinct outcomes are coarse-grained.

Physical units belong to the outcome scale. If $A$ measures energy, its spectral values and expectation values have energy units. The projectors are dimensionless. An affine recalibration

A'=cA+dI

changes the reported outcome values to $a'=ca+d$ without changing the spectral subspaces when $c\ne0$ .

Compatibility

For finite-dimensional sharp observables, commuting operators,

[A,B]=0,

admit a common spectral refinement and can be represented in a simultaneous eigenbasis. Degeneracy must be handled with joint eigenspaces rather than by matching arbitrary eigenvectors.

For general POVMs, joint measurability is broader than pairwise operator commutation of individual effects. Compatible Observables is the canonical treatment.

Minimal Example

For a spin- $1/2$ component along a unit vector $\mathbf n$ ,

S_{\mathbf n} =\frac{\hbar}{2}\, \mathbf n\cdot\boldsymbol{\sigma}.

Its outcomes are $\pm\hbar/2$ , with projectors

\Pi_\pm =\frac12 \left( I\pm\mathbf n\cdot\boldsymbol{\sigma} \right).

For a state

\rho=\frac12 \left( I+\mathbf r\cdot\boldsymbol{\sigma} \right),

the Born probabilities are

p_\pm =\operatorname{Tr}(\rho\Pi_\pm) =\frac12 \left( 1\pm\mathbf r\cdot\mathbf n \right).

The observable specifies two outcome values and their projectors. The state determines the numerical probability distribution.

Common Aliases

physical quantity;
dynamical variable, especially in older texts;
sharp observable, when a self-adjoint operator or PVM is intended;
generalized observable, when a POVM is intended.

The words operator and observable are often used interchangeably in elementary contexts, but not every linear operator represents an observable.

Common Confusions

A Hermitian matrix is the finite-dimensional form of a sharp observable; infinite-dimensional observables require domain care.
Not every linear operator represents an observable.
Applying an operator to a ket is an algebraic action, not a physical measurement process.
Measurement outcomes and expectation values are different objects.
The spectrum is the set of possible sharp values, not the probability distribution; probabilities also require a state.
Degenerate outcomes correspond to subspaces, not arbitrarily chosen eigenvectors.
A POVM determines probabilities but not a unique state-update rule.
A self-adjoint first-moment operator does not, in general, uniquely determine an entire POVM.
An observable is not identical to the apparatus that implements it.
Noncommutation of sharp observables is related to incompatibility, but generalized joint measurability requires the POVM framework.

References

J. J. Sakurai and J. Napolitano, Modern Quantum Mechanics, 3rd ed., Cambridge University Press, 2020, ch. 1.
R. Shankar, Principles of Quantum Mechanics, 2nd ed., Springer, 1994, chs. 4 and 5.
P. Busch, P. J. Lahti, and P. Mittelstaedt, The Quantum Theory of Measurement, 2nd rev. ed., Springer, 1996.
B. C. Hall, Quantum Theory for Mathematicians, Springer, 2013, chs. 7 and 10.