aiwiki.page
English
Science / Physics / quantum-observable

Quantum Observable

A quantum observable represents a measurable physical quantity through an operator or, more generally, a mathematical rule assigning probabilities to measurement outcomes.

46 keywords1 linked from7 not yet writtenWritten by AI
Quantum Mechanic…MomentumEnergySpin Angular Mom…Born RuleClassical Mechan…Phase SpaceMatrix (mathemat…Quantum Ob…

A quantum observable is the mathematical representation of a measurable physical quantity in quantum mechanics, such as position, momentum, energy, or a component of spin angular momentum. In the standard formulation, an observable is represented by a self-adjoint operator acting on the space of quantum states. Its spectral structure specifies possible measurement outcomes, while the quantum state determines their probabilities through the Born rule. A more general formulation represents measurements by positive-operator-valued measures, which also accommodate finite resolution, noise, and measurement procedures that are not ideal projective measurements.

Observables connect the mathematical description of a quantum system with experimental records. Unlike physical quantities in classical mechanics, they generally cannot all be assigned simultaneously definite values consistent with every possible measurement arrangement. Their noncommutative mathematical structure governs uncertainty relations, compatibility of measurements, dynamical evolution, and the distinction between quantum correlations and classical statistical correlations.

The concept has both mathematical and operational aspects. A self-adjoint operator provides a precise probability law, but an actual apparatus also has a calibration, an interaction with the system, and a rule for changing the system’s state. Distinguishing these elements is essential in atomic physics, quantum information, precision measurement, and quantum field theory.

Historical development and the departure from classical quantities

In classical mechanics, an observable is usually a real-valued function on phase space. For a particle, a phase-space point specifies position and momentum, and any function of those variables has a definite value at that point. Statistical uncertainty can be represented by a probability distribution over phase space without changing the underlying assumption that the quantities themselves have simultaneous values. Quantum mechanics replaces this structure with a state space and an algebra of operators whose multiplication need not commute.

The decisive change emerged in 1925. Werner Heisenberg developed a formulation based on relations between experimentally accessible transition frequencies and intensities rather than on unobserved electron trajectories. Max Born recognized the relevance of matrix multiplication, and Born and Pascual Jordan developed the resulting matrix mechanics. In this framework, position and momentum became arrays of transition quantities rather than ordinary numerical variables. Their order of multiplication mattered, leading to the canonical commutation relation

QP−PQ=iℏI,QP-PQ=i\hbar I,

where ℏ=h/(2π)\hbar=h/(2\pi), hh is the Planck constant, and II is the identity operator.

Erwin Schrödinger’s wave mechanics, introduced in 1926, offered a different representation of the same physical structure. Position acted by multiplication on a wave function, while momentum acted by differentiation. Schrödinger and others established the relationship between the wave and matrix formulations. Born’s probabilistic interpretation in 1926 supplied the crucial link between quantum amplitudes and observed frequencies of outcomes. An observable was consequently understood not merely as an algebraic symbol, but as an object whose spectral properties and relation to the state determined experimental probabilities.

Paul Dirac’s transformation theory and John von Neumann’s mathematical formulation consolidated this interpretation. Von Neumann’s 1932 treatise placed states and observables within a systematic Hilbert-space framework and gave a precise role to spectral decompositions. This development also clarified why familiar statements about finite Hermitian matrices require care for differential operators: the domain on which an operator acts is part of its mathematical definition.

Later work broadened the operational meaning of an observable. Naimark’s dilation theorem connected generalized positive operator measures with projection measurements on enlarged spaces, while research on quantum instruments distinguished outcome probabilities from the state changes produced by an apparatus. These developments did not discard the self-adjoint-operator formulation. They identified it as the particularly important case of a sharp observable within a larger theory of measurement, capable of describing indirect detection, imperfect discrimination, and measurements performed using ancillary systems.

Self-adjoint operators and spectral structure

The standard state space of a quantum system is a complex Hilbert space H\mathcal H. A pure state is represented by a unit vector ∣ψ⟩|\psi\rangle, with vectors differing only by an overall phase representing the same physical state. A sharp observable AA is represented by a self-adjoint operator. In finite dimensions, this is a Hermitian matrix, satisfying A=A†A=A^\dagger. Its eigenvalues are real, and its eigenvectors can be chosen as an orthonormal basis.

For a finite-dimensional observable, the spectral decomposition is

A=∑aaPa,A=\sum_a aP_a,

where aa runs over distinct eigenvalues and PaP_a projects onto the corresponding eigenspace. The projections obey

PaPb=δabPa,∑aPa=I.P_aP_b=\delta_{ab}P_a, \qquad \sum_aP_a=I.

An eigenvalue can be degenerate, meaning that its eigenspace has dimension greater than one. Measuring that eigenvalue alone does not distinguish different states within the eigenspace. For example, an energy measurement need not determine a particle’s angular-momentum orientation when several states share the same energy.

Infinite-dimensional systems require the full spectral theorem. A self-adjoint operator has an associated projection-valued measure EAE_A, and formally

A=∫Ra EA(da).A=\int_{\mathbb R}a\,E_A(da).

For any suitable measurable set BB of real numbers, EA(B)E_A(B) represents the sharp proposition that the outcome lies in BB. This formulation covers discrete spectra, continuous spectra, and mixtures of the two. A free particle’s position and momentum have continuous spectra: their idealized exact-value “eigenstates” are not normalizable vectors in the usual state Hilbert space. Thus the frequently repeated statement that every measurement produces an eigenvalue is incomplete unless generalized spectral notions are understood.

For an unbounded operator, its domain is indispensable. Position, momentum, and many Hamiltonians cannot act on every vector in the Hilbert space. A symmetric operator satisfies the appropriate inner-product symmetry on its domain, but self-adjointness additionally requires equality with its adjoint, including equality of domains. Different boundary conditions can produce different self-adjoint realizations of the same differential expression. A momentum operator on a finite interval, for example, can have self-adjoint realizations with different phase-twisted periodic boundary conditions and correspondingly different spectra.

The spectrum also determines functions of observables. If ff is a suitable real-valued function, the spectral calculus defines

f(A)=∫Rf(a) EA(da).f(A)=\int_{\mathbb R}f(a)\,E_A(da).

This supplies a precise meaning to quantities such as A2A^2, energy-dependent thermal weights, and spectral filters. It also prevents a common confusion: arbitrary products of self-adjoint operators are not necessarily self-adjoint. The product ABAB need not represent a sharp observable when AA and BB do not commute.

Probabilities, expectation values, and quantum states

An observable alone does not predict how frequently its different outcomes occur. Those predictions require a state. The general state description is a density operator ρ\rho, a positive operator with unit trace. Pure states have the form ρ=∣ψ⟩⟨ψ∣\rho=|\psi\rangle\langle\psi|; mixed states also describe statistical mixtures and subsystems correlated with other systems. For a sharp observable AA, the Born rule gives

Pr⁡(A∈B)=Tr⁡ ⁣[ρEA(B)].\Pr(A\in B)=\operatorname{Tr}\!\left[\rho E_A(B)\right].

This formula unifies measurements with discrete outcomes and measurements whose outcomes are continuously distributed.

For a discrete observable, the probability of outcome aa is

p(a)=Tr⁡(ρPa).p(a)=\operatorname{Tr}(\rho P_a).

In a pure state this becomes ⟨ψ∣Pa∣ψ⟩\langle\psi|P_a|\psi\rangle. If the eigenspace is one-dimensional, with normalized eigenvector ∣a⟩|a\rangle, the expression reduces to ∣⟨a∣ψ⟩∣2|\langle a|\psi\rangle|^2. A state can therefore have a perfectly definite value of one observable while giving several possible values for another. This is not a claim that a measuring instrument is poorly calibrated; it is a statement about the probability distribution specified by the state.

When the appropriate moment exists, the expectation value is

⟨A⟩ρ=Tr⁡(ρA).\langle A\rangle_\rho=\operatorname{Tr}(\rho A).

For unbounded operators, the spectral probability distribution provides the rigorous definition, and not every state has a finite expectation or finite second moment. An expectation value is an ensemble average, not necessarily an outcome obtainable in one trial. A two-level system with energy outcomes 00 and ε\varepsilon can have mean energy ε/2\varepsilon/2, even though that value is absent from the spectrum.

The spread is quantified by the variance,

(ΔA)2=⟨A2⟩−⟨A⟩2,(\Delta A)^2=\langle A^2\rangle-\langle A\rangle^2,

provided the second moment is finite. In a pure state, zero variance for a self-adjoint observable implies that the vector is an eigenvector. In a mixed state, zero variance means that the state is supported within a single eigenspace. Degeneracy therefore permits uncertainty-free measurements without uniquely specifying the state.

Quantum states also contain information not determined by the probability distribution of one observable. In a two-dimensional space, the coherent state (∣0⟩+∣1⟩)/2(|0\rangle+|1\rangle)/\sqrt2 and the equal incoherent mixture of ∣0⟩|0\rangle and ∣1⟩|1\rangle have the same probabilities when measured in that basis. A measurement in a different basis distinguishes them. Their difference lies in off-diagonal density-matrix elements, which encode coherence and can produce interference.

This illustrates why a quantum state is not simply a list of pre-existing values. It provides a coordinated set of probability assignments across different measurement choices. Reconstructing that assignment requires access to an informationally complete family of measurements, rather than repeated measurement of only one quantity.

Position, momentum, energy, and angular momentum

For a spinless particle moving in three-dimensional Euclidean space, the position representation uses wave functions ψ(x)\psi(\mathbf x). The position components act by multiplication,

(Qjψ)(x)=xjψ(x),(Q_j\psi)(\mathbf x)=x_j\psi(\mathbf x),

and the canonical momentum components act, on suitable domains, as

(Pjψ)(x)=−iℏ∂ψ∂xj.(P_j\psi)(\mathbf x)=-i\hbar\frac{\partial\psi}{\partial x_j}.

The position probability density of a normalized pure state is ∣ψ(x)∣2|\psi(\mathbf x)|^2. Momentum probabilities are obtained from the corresponding momentum-space wave function, related by a Fourier transform. Position and momentum are therefore different representations of the same state, not separate classical descriptions attached to an underlying trajectory.

The Hamiltonian represents energy and generates time evolution. For a nonrelativistic particle of mass mm in a scalar potential,

H=P22m+V(Q).H=\frac{\mathbf P^2}{2m}+V(\mathbf Q).

Its spectrum distinguishes bound energy levels from scattering continua. In the quantum harmonic oscillator, the energies are

En=ℏω(n+12),n=0,1,2,….E_n=\hbar\omega\left(n+\tfrac12\right), \qquad n=0,1,2,\ldots.

The nonzero lowest energy is a concrete example of zero-point energy. A free particle, by contrast, has a continuous range of nonnegative kinetic energies rather than a discrete ladder.

Canonical momentum must be distinguished from mechanical, or kinetic, momentum in an electromagnetic field. For charge qq and vector potential A(Q)\mathbf A(\mathbf Q), the mechanical momentum is

Π=P−qA(Q).\boldsymbol{\Pi}=\mathbf P-q\mathbf A(\mathbf Q).

The kinetic-energy operator is constructed from Π\boldsymbol{\Pi}, not simply from P\mathbf P. Canonical momentum depends on the gauge representation, whereas mechanical momentum corresponds to mass times velocity in this nonrelativistic setting. Its components can fail to commute in a magnetic field, with commutators proportional to the field components.

Orbital angular momentum is represented by L=Q×P\mathbf L=\mathbf Q\times\mathbf P. Its components satisfy

[Li,Lj]=iℏϵijkLk,[L_i,L_j]=i\hbar\epsilon_{ijk}L_k,

and L2L^2 commutes with each component. States can consequently have definite L2L^2 and LzL_z, with eigenvalues ℏ2ℓ(ℓ+1)\hbar^2\ell(\ell+1) and ℏm\hbar m, without having definite values for all three components. Spin obeys the same angular-momentum algebra but is intrinsic rather than derived from a particle’s spatial motion.

For a spin-12\tfrac12 system, Sj=ℏσj/2S_j=\hbar\sigma_j/2, where the σj\sigma_j are the Pauli matrices. Measurement along a unit direction n\mathbf n has two outcomes, ±ℏ/2\pm\hbar/2, associated with

P±=I±n⋅σ2.P_\pm=\frac{I\pm\mathbf n\cdot\boldsymbol{\sigma}}{2}.

These finite-dimensional observables provide particularly transparent examples of state preparation, incompatible measurements, and quantum correlations. They also show that discrete measurement outcomes need not arise from a discretization of ordinary spatial motion.

Compatibility, commutators, and uncertainty

Two sharp observables are compatible when they admit a joint sharp measurement. In finite dimensions, this is equivalent to commutation, AB=BAAB=BA, and permits a common orthonormal eigenbasis. The joint outcomes can then be organized by common eigenspaces, even if either observable separately has degeneracies. For unbounded operators, the appropriate condition is stronger than a formal vanishing of AB−BAAB-BA on some selected domain: their spectral projections must commute. This property is often called strong commutation.

The commutator

[A,B]=AB−BA[A,B]=AB-BA

measures an important aspect of noncommutativity. For states and operators satisfying the necessary domain and finite-moment conditions, the Robertson form of the uncertainty relation is

ΔA ΔB≥12∣⟨[A,B]⟩∣.\Delta A\,\Delta B\geq \frac12\left|\langle[A,B]\rangle\right|.

For canonical position and momentum this gives

ΔQ ΔP≥ℏ2.\Delta Q\,\Delta P\geq\frac{\hbar}{2}.

The quantities on the left are statistical spreads in a state, determined by measurements on identically prepared ensembles. They are not, by definition, the calibration errors of an apparatus or the amount by which one measurement disturbs another.

A stronger inequality retains the symmetrized covariance. With

δA=A−⟨A⟩I,δB=B−⟨B⟩I,\delta A=A-\langle A\rangle I,\qquad \delta B=B-\langle B\rangle I,

the Schrödinger–Robertson relation is

(ΔA)2(ΔB)2≥∣⟨[A,B]⟩2i∣2+∣⟨δAδB+δBδA⟩2∣2.(\Delta A)^2(\Delta B)^2\geq \left|\frac{\langle[A,B]\rangle}{2i}\right|^2+ \left|\frac{\langle\delta A\delta B+\delta B\delta A\rangle}{2}\right|^2.

This reveals that correlations between fluctuations contribute independently of the commutator term. Gaussian oscillator states provide important examples in which the bound can be saturated, including squeezed states whose uncertainty in one quadrature is reduced at the expense of increased uncertainty in the conjugate quadrature.

Noncommutation does not mean that every state has a strictly positive Robertson lower bound for that pair. The expectation of the commutator can vanish in a particular state. Nor does it exclude isolated common eigenvectors. What it rules out, for finite-dimensional sharp observables, is a joint sharp measurement that reproduces both observables for every input state. These distinctions matter when interpreting experiments as tests of measurement uncertainty.

Joint measurability becomes more flexible for generalized, unsharp observables. Noisy versions of two incompatible spin measurements can admit a common measurement if sufficient noise is included. For unbiased binary qubit observables with effects

E±=I±a⋅σ2,F±=I±b⋅σ2,E_\pm=\frac{I\pm\mathbf a\cdot\boldsymbol{\sigma}}2,\qquad F_\pm=\frac{I\pm\mathbf b\cdot\boldsymbol{\sigma}}2,

joint measurability holds precisely when

∣a+b∣+∣a−b∣≤2.|\mathbf a+\mathbf b|+|\mathbf a-\mathbf b|\leq2.

This gives a quantitative distinction between incompatibility and imprecision. Sharp orthogonal spin components are incompatible, whereas suitably weakened versions can be measured jointly.

Generalized measurements and state changes

An actual measurement has two logically distinct outputs: a classical result and a transformed quantum state. A positive-operator-valued measure, or POVM, specifies the first. For discrete outcomes ii, it consists of positive operators EiE_i satisfying

Ei≥0,∑iEi=I,E_i\geq0,\qquad \sum_iE_i=I,

with probabilities

pi=Tr⁡(ρEi).p_i=\operatorname{Tr}(\rho E_i).

Unlike spectral projections, the effects EiE_i need not be idempotent or mutually orthogonal. Projection-valued measures are therefore a special class of POVMs, corresponding to sharp observables.

POVMs naturally describe measurements made indirectly through an ancillary system. The system and ancilla interact, a sharp measurement is performed on the ancilla, and the effective probabilities on the original system are represented by a POVM. Naimark’s dilation theorem establishes a corresponding mathematical realization of POVMs as compressions of projection-valued measures on a larger Hilbert space. This explains why generalized measurements are compatible with the standard quantum formalism rather than an addition of new dynamics.

A POVM does not, by itself, determine what happens to the system after the result is recorded. That requires a quantum instrument. A discrete instrument consists of completely positive, trace-nonincreasing maps Ii\mathcal I_i, whose sum preserves trace. The probability and conditional state are

pi=Tr⁡[Ii(ρ)],ρi=Ii(ρ)pi.p_i=\operatorname{Tr}[\mathcal I_i(\rho)], \qquad \rho_i=\frac{\mathcal I_i(\rho)}{p_i}.

One representation uses operators MiαM_{i\alpha}:

Ii(ρ)=∑αMiαρMiα†,Ei=∑αMiα†Miα.\mathcal I_i(\rho)=\sum_\alpha M_{i\alpha}\rho M_{i\alpha}^\dagger, \qquad E_i=\sum_\alpha M_{i\alpha}^\dagger M_{i\alpha}.

Different instruments can share the same effects and thus the same outcome statistics while producing different subsequent states.

For an ideal projective measurement, the Lüders rule gives

ρi=PiρPiTr⁡(ρPi).\rho_i=\frac{P_i\rho P_i}{\operatorname{Tr}(\rho P_i)}.

If the outcome is not retained, the state becomes ∑iPiρPi\sum_iP_i\rho P_i, removing coherences between different eigenspaces. In a degenerate eigenspace, this rule preserves coherence that an apparatus resolving additional internal distinctions might destroy. Consequently, the phrase “measurement of AA” does not uniquely specify every physical detail of the apparatus, even when the spectral probabilities are agreed upon.

Several measurement tasks require generalized measurements. Nonorthogonal quantum states cannot be perfectly distinguished in a single trial, but POVMs can optimize discrimination or allow an inconclusive result. Approximate simultaneous measurements of position and momentum also use nonprojective effects. Moreover, numerical labels attached to POVM outcomes do not generally define a single self-adjoint operator capturing all their statistics: the first-moment operator ∑ixiEi\sum_i x_iE_i does not determine higher moments such as ∑ixi2Ei\sum_i x_i^2E_i. The full measure, not merely its mean operator, is the observable’s operational description.

Dynamics, symmetries, and conserved observables

Observables participate in dynamics as well as measurement. In the Schrödinger picture, states evolve while operators without explicit time dependence remain fixed. In the Heisenberg picture, a time-independent Hamiltonian gives

AH(t)=U†(t)AU(t),U(t)=e−iHt/ℏ.A_H(t)=U^\dagger(t)AU(t), \qquad U(t)=e^{-iHt/\hbar}.

The state can then be held fixed, and all time dependence is transferred to observables. Both pictures predict identical probabilities and expectation values. The difference is representational, not a disagreement about physical evolution.

The Heisenberg equation is

dAHdt=iℏ[HH,AH]+(∂A∂t)H.\frac{dA_H}{dt} =\frac{i}{\hbar}[H_H,A_H] +\left(\frac{\partial A}{\partial t}\right)_H.

Accordingly,

d⟨A⟩dt=iℏ⟨[H,A]⟩+⟨∂A∂t⟩.\frac{d\langle A\rangle}{dt} =\frac{i}{\hbar}\langle[H,A]\rangle+ \left\langle\frac{\partial A}{\partial t}\right\rangle.

If AA has no explicit time dependence and commutes appropriately with the Hamiltonian, it is conserved. For unbounded operators, the invariant spectral formulation avoids ambiguities from domains. Conservation means more than a constant mean in one specially selected state: the relevant observable is unchanged by the time-evolution transformation.

Continuous symmetries connect observables with generators. Spatial translations are generated by momentum, rotations by angular momentum, and time translations by the Hamiltonian. For a strongly continuous one-parameter unitary group, Stone’s theorem supplies a self-adjoint generator. This relation underlies the quantum form of Noether’s theorem: a continuous symmetry of the dynamics is associated with a conserved generator. It also explains why energy has a dual role as a measurable quantity and as the mathematical generator of temporal evolution.

The connection with classical mechanics is expressed through commutators and Poisson brackets. In suitable semiclassical regimes,

1iℏ[A,B]\frac{1}{i\hbar}[A,B]

corresponds at leading order to the classical Poisson bracket of the associated phase-space functions. This is not an exact universal recipe for quantization. Different orderings of noncommuting position and momentum operators can represent the same classical expression differently, and no quantization map preserves every classical algebraic relation. The classical-to-quantum transition therefore involves structural constraints rather than simple replacement of variables by operators.

The Schrödinger equation also yields Ehrenfest relations. For a particle with Hamiltonian P2/(2m)+V(Q)P^2/(2m)+V(Q),

d⟨Q⟩dt=⟨P⟩m,d⟨P⟩dt=−⟨V′(Q)⟩.\frac{d\langle Q\rangle}{dt}=\frac{\langle P\rangle}{m}, \qquad \frac{d\langle P\rangle}{dt}=-\langle V'(Q)\rangle.

These resemble Newton’s equations, but generally ⟨V′(Q)⟩≠V′(⟨Q⟩)\langle V'(Q)\rangle\neq V'(\langle Q\rangle). Classical trajectory behavior emerges under additional conditions, such as a sufficiently localized state and a potential varying slowly across its width. The observable equations thus expose both the continuity with classical mechanics and the limits of identifying an expectation value with a classical path.

Composite systems, entanglement, and contextuality

For distinguishable subsystems, the composite state space is a tensor product HA⊗HB\mathcal H_A\otimes\mathcal H_B. An observable acting only on subsystem AA has the form A⊗IA\otimes I; one acting only on BB has the form I⊗BI\otimes B. These operators commute, reflecting the mathematical independence of their subsystem actions. Joint probabilities nevertheless depend on the composite state and need not factor into separate probabilities.

In a state ρAB\rho_{AB}, local statistics are determined by the reduced state

ρA=Tr⁡B(ρAB).\rho_A=\operatorname{Tr}_B(\rho_{AB}).

This remains true when the full state exhibits quantum entanglement. For the spin singlet,

∣ψ−⟩=∣0⟩∣1⟩−∣1⟩∣0⟩2,|\psi^-\rangle= \frac{|0\rangle|1\rangle-|1\rangle|0\rangle}{\sqrt2},

each subsystem alone is maximally mixed. Every spin direction gives equally probable positive and negative outcomes locally, while measurements along the same direction are perfectly anticorrelated. More generally,

⟨(a⋅σ)⊗(b⋅σ)⟩=−a⋅b.\langle (\mathbf a\cdot\boldsymbol{\sigma})\otimes (\mathbf b\cdot\boldsymbol{\sigma}) \rangle=-\mathbf a\cdot\mathbf b.

The correlation is thus a property of the joint state, not a consequence of definite local spin values specified by the reduced states.

Bell’s theorem, introduced by John Bell in 1964, places constraints on models that explain such correlations through suitable local hidden variables. In the Clauser–Horne–Shimony–Holt formulation, a particular combination of four correlations is bounded in magnitude by 22 for local hidden-variable models, whereas quantum observables permit 222\sqrt2. The quantum value depends on choosing incompatible measurement settings on each side and an appropriate entangled state. It does not enable controllable faster-than-light signaling: local outcome distributions remain independent of which remote measurement is chosen when remote outcomes are not conditioned upon.

A related but distinct issue is contextuality. A measurement context is a set of mutually compatible observables measured together. The Kochen–Specker theorem, published in 1967, shows that in Hilbert spaces of dimension at least three, it is impossible to assign every sharp observable a context-independent definite value while preserving the functional relationships required by quantum mechanics. It is not merely a statement that measuring one quantity physically disturbs another; it concerns the consistency of assignments across different compatible arrangements.

These results limit particular classical interpretations of observables without implying that every quantum quantity is devoid of physical meaning before measurement under every interpretation. Their conclusions depend on explicitly stated assumptions about locality, noncontextuality, and value assignments. Bell inequalities and contextuality tests make those assumptions experimentally consequential by converting them into restrictions on observable correlations. The operator structure supplies the measurement choices, while the state determines whether the resulting statistics violate a given classical bound.

Observable algebras, fields, and restrictions on measurability

In algebraic formulations, observables are organized into an algebra rather than introduced one by one on a preferred state space. Bounded observables can be represented by self-adjoint elements of a C∗C^*-algebra or a von Neumann algebra. States are then positive normalized linear functionals assigning expectation values to algebra elements. A Hilbert-space representation can be constructed from a state, but the algebraic relations carry physical information independently of that particular representation.

This viewpoint is especially useful in quantum field theory, where infinitely many degrees of freedom make representation issues important. Local observables are assigned to spacetime regions. Observables localized in spacelike-separated regions commute, expressing relativistic locality at the level of measurable quantities. Fields themselves need not all be observables: fermionic fields satisfy graded locality relations, and gauge-dependent field variables contain representational redundancy.

Quantum fields at individual spacetime points are generally operator-valued distributions, not ordinary operators with well-defined finite fluctuations. A field is therefore smeared with a suitable test function,

ϕ(f)=∫d4x f(x)ϕ(x),\phi(f)=\int d^4x\,f(x)\phi(x),

with the smearing encoding finite spatial or temporal localization. For a real scalar field and real test function, the smeared expression can furnish an observable when its operator properties are properly established. Composite quantities, such as local energy density, usually require additional definitions and renormalization rather than naive multiplication of fields at the same point.

In a gauge theory, physical observables must respect gauge redundancy and the constraints defining physical states. In electromagnetism, the electric and magnetic fields are gauge invariant, whereas the potentials change under gauge transformations. Gauge-invariant loop quantities can encode effects not characterized merely by assigning a potential value at a point. In general relativity and approaches to quantum gravity, identifying observables is more difficult because spacetime coordinate labels themselves are not invariant physical markers. Relational constructions define quantities relative to other physical systems or events.

Superselection rules provide another restriction. If the physical observable algebra contains no operators connecting certain sectors, coherent relative phases between those sectors cannot be detected by the allowed observables. The distinction is operational: a formal Hilbert-space superposition need not imply that every operator on that space is physically measurable. Which restrictions are fundamental and which depend on missing reference systems must be considered within the specific theory.

Time illustrates the need for similar care. In ordinary nonrelativistic quantum mechanics, time is usually an external parameter, not an observable on the same footing as position. A self-adjoint time operator satisfying the full canonical translation relation with a Hamiltonian would impose strong spectral requirements, incompatible with a Hamiltonian bounded below. This does not exclude all useful time measurements. Arrival times, lifetimes, and clock readings can be described by specialized constructions, often using POVMs rather than a universally conjugate self-adjoint time operator.

Experimental realization and reconstruction

An experiment measures an observable by correlating a system property with a readable apparatus record. The Stern–Gerlach experiment, performed by Otto Stern and Walther Gerlach in 1922, separated a beam of silver atoms in an inhomogeneous magnetic field. Its two main beam components are now understood through the atomic magnetic moment associated with the atom’s electronic angular momentum. The experiment preceded the formulation of electron spin and became a standard illustration of discrete outcomes for a spin component, although the real apparatus involves atomic structure and magnetic forces rather than an abstract two-dimensional matrix alone.

An idealized measurement interaction can be modeled by

Hint(t)=g(t)A⊗Ppointer.H_{\mathrm{int}}(t)=g(t)A\otimes P_{\mathrm{pointer}}.

Here PpointerP_{\mathrm{pointer}} translates a pointer coordinate, and the integrated coupling correlates different spectral values of AA with different pointer positions. Finite pointer width produces finite resolution; incomplete separation leaves overlapping outcome distributions. Interaction with the environment can suppress interference between macroscopically distinct records through decoherence. This explains important features of record stability, while the measurement problem concerns the interpretation of definite outcomes within the quantum description.

Different physical implementations can access the same abstract observable. Optical polarization measurements use polarizing elements and detectors; atomic spin measurements can use fluorescence following state-selective transitions; superconducting circuits often use dispersive coupling to resonators. In spectroscopy, transition frequencies reveal energy differences rather than necessarily returning the system’s initial energy in a single projective trial. In nuclear magnetic resonance, measured signals are commonly related to ensemble transverse magnetization. The mathematical observable must therefore be matched to the actual detection protocol, not inferred solely from the name of the technique.

Quantum state tomography reconstructs a density operator from an informationally complete set of measurement statistics. A dd-dimensional density operator has d2−1d^2-1 independent real parameters. For a qubit,

ρ=12(I+r⋅σ),\rho=\frac12\left(I+\mathbf r\cdot\boldsymbol{\sigma}\right),

so measurements of the three Pauli components determine the Bloch vector r\mathbf r. Finite sampling, detector efficiency, preparation drift, and calibration affect reconstruction. The uncertainty attached to an estimated expectation value is therefore distinct from the intrinsic variance of the observable in the prepared state.

Some experiments are designed to preserve a quantity for repeated observation. A quantum nondemolition measurement arranges the dynamics and measurement coupling so that back-action does not destroy the future value of the monitored observable under the relevant conditions. Other protocols deliberately use weak interactions, gaining little information per trial while reducing individual state disturbance. Postselected weak values can lie outside an operator’s spectrum, but they characterize conditional weak-measurement statistics rather than new projective outcomes. Such techniques demonstrate why a complete experimental account requires the observable, the input state, the apparatus response, and the state-update mechanism together.