Canonical Quantization: Algebra, Representation, and State

Canonical quantization is a controlled construction for selected classical observables, not a rule that turns every classical function into an operator. For unconstrained bosonic data on a fixed Cauchy slice, it first replaces the Poisson relations of elementary observables by defining relations of a quantum algebra. A Hilbert-space representation and a state are separate choices. This page makes those choices explicit and tests the construction on finitely many modes of a free real scalar field; nonlinear ordering, constrained systems, and the continuum limit mark the boundary of the method developed here.

Helpful background. Hamiltonian initial data and phase space supplies canonical fields and their functional Poisson brackets. Commutators and operator exponentials supplies the algebraic identities used below, while symplectic forms and Hamiltonian flows gives the coordinate-free version. Unbounded operators, domains, closure, and adjoints becomes important once abstract generators are represented as operators.

From canonical data to the CCR algebra

Begin with a real classical phase space and choose a real linear space of elementary observables that, after adjoining constants, is closed under the Poisson bracket. For one canonical pair, the defining classical relation is

$$\{q,p\}=1.$$

Canonical quantization introduces abstract generators $Q=Q^*$ and $P=P^*$ and imposes

$$[Q,P]=i\mathbf 1.$$

The word abstract matters: at this stage $Q$ and $P$ are elements of a unital $*$-algebra, not yet operators on a Hilbert space. The relation fixes their noncommutativity but does not choose spectra, domains, a vacuum, or expectation values Schwartz 2014, § 2.3, pp. 20–26; Hollands and Wald 2015, § 1, pp. 4–8 and § 2.1, pp. 9–12, Open PDF.

For a real scalar field at a fixed time $t_0$, let $f,g\in C_c^\infty(\mathbb R^3,\mathbb R)$ and introduce smeared generators

$$\Phi(f)=\int \mathrm d^3\mathbf x\,f(\mathbf x)\Phi(\mathbf x), \qquad \Pi(g)=\int \mathrm d^3\mathbf x\,g(\mathbf x)\Pi(\mathbf x).$$

The equal-time canonical commutation relations, or CCR, are

$$\begin{aligned} [\Phi(f),\Phi(g)]&=0,\\ [\Pi(f),\Pi(g)]&=0,\\ [\Phi(f),\Pi(g)]&= i\int \mathrm d^3\mathbf x\,f(\mathbf x)g(\mathbf x)\,\mathbf 1. \end{aligned}$$

For complex test functions the generators extend complex-linearly and obey $\Phi(f)^*=\Phi(\overline f)$ and $\Pi(g)^*=\Pi(\overline g)$. The familiar expression

$$[\Phi(\mathbf x),\Pi(\mathbf y)] =i\delta^{(3)}(\mathbf x-\mathbf y)\mathbf 1$$

is distributional shorthand for the smeared relation, not an equation between operators at individual points. Smearing is what gives the delta distribution a defined action Schwartz 2014, § 2.3, pp. 20–26. The fuller spacetime and domain story is developed in quantum fields as operator-valued distributions.

These are relations for bosonic, or even, canonical variables. Fermionic canonical variables are odd and instead lead to canonical anticommutation relations, or CAR, such as

$$\{b_r,b_s^*\}=\delta_{rs}\mathbf 1, \qquad \{b_r,b_s\}=0.$$

Replacing CCR by CAR is a change of graded algebra, not an optional sign convention. CAR are included here only to orient the choice of algebra; the canonical quantization of the free Dirac field develops the fermionic construction.

Why the elementary observables must be chosen

The prescription is unambiguous on the selected linear generators, but it does not determine a quantization map for every nonlinear classical observable. Classically $qp=pq$, whereas the quantum relation gives

$$QP-PQ=i\mathbf 1.$$

Thus $QP$, $PQ$, and $(QP+PQ)/2$ cannot all be the same image of the classical product $qp$. An ordering rule can choose among them, but that rule is additional data. Canonical quantization should therefore state which observables generate the algebra and which composite observables have actually been defined.

A polynomial CCR algebra keeps the generators visible, but its represented generators will generally be unbounded. To fix the sign in the bounded Weyl presentation, let

$$\begin{aligned} R(F)&=fQ+gP,\\ \sigma(F,G)&=fg'-gf',\\ [R(F),R(G)]&=i\sigma(F,G)\mathbf 1. \end{aligned}$$

for $F=(f,g)$ and $G=(f',g')$. The construction replaces $R(F)$ by unitary generators $W(F)$, formally corresponding to $e^{iR(F)}$, and takes

$$\begin{aligned} W(F)W(G) &=e^{-i\sigma(F,G)/2}W(F+G),\\ W(F)^*&=W(-F),\\ W(0)&=\mathbf 1. \end{aligned}$$

as the defining relations. For a field, $F$ is a smeared phase-space vector and $\sigma$ is the corresponding classical symplectic form. The Weyl form packages the same linear canonical data without treating $Q$ and $P$ themselves as bounded elements. Its theorem-level completion leads toward quasilocal C*-algebras and inductive limits.

Algebra, representation, and state

Three objects answer different questions and should not be collapsed into one another.

Object	Role and limit
Algebra $\mathcal A$	Provides: generators, products, adjoints, and commutation or anticommutation relations. Does not select: a Hilbert space, operator domains, or expectation values.
Representation $\rho$	Provides: a $*$-preserving operator realization on a Hilbert space and common domain. Does not select: a preferred state, or establish self-adjointness and spectral claims without further analysis.
State $\omega$	Provides: positive normalized expectation values and correlation functions. Does not select: the elementary algebra, or make every unbounded product well-defined.

Positivity and normalization mean

$$\omega(A^*A)\geq 0, \qquad \omega(\mathbf 1)=1.$$

In a bounded representation, a normalized vector $\Psi\in\mathcal H$ defines the vector state $\omega_\Psi(A)=\langle\Psi,\rho(A)\Psi\rangle$. Likewise, a density operator $D$ gives $\omega_D(A)=\operatorname{Tr}(D\rho(A))$ when the trace is defined. For a polynomial CCR representation, such formulas require vectors in a common invariant domain and matrix elements that exist there.

Conversely, the Gelfand–Naimark–Segal construction starts from an abstract state and produces a cyclic triple $(\mathcal H_\omega,\rho_\omega,\Omega_\omega)$ satisfying

$$\omega(A)= \langle\Omega_\omega, \rho_\omega(A)\Omega_\omega\rangle.$$

For a general unital $*$-algebra, $\rho_\omega(A)$ acts first on the dense quotient domain obtained by removing the state’s null vectors; only in the C*-algebra case does the representation automatically extend by bounded operators to all of $\mathcal H_\omega$. The construction, its hypotheses, and the unitary uniqueness of the cyclic triple are summarized by Hollands and Wald 2015, § 2.1, pp. 12–14, Open PDF. The CCR still need not select the state, and field theories can have inequivalent representations. See vacua, states, and representations for the broader comparison and states, GNS representations, and folia for the theorem-first treatment.

For the polynomial algebra, the operators $\rho(Q)$ and $\rho(P)$ are generally unbounded. Their commutator is therefore asserted on a specified common invariant dense domain, not automatically on every vector in $\mathcal H$. That analytic qualification enters at the representation layer; it is not supplied by writing down the abstract CCR.

A reproducible canonical-quantization workflow

The method can be organized as a sequence whose output and stopping point are explicit.

1. Specify the classical input. Give the phase space, symplectic or Poisson structure, boundary conditions, constraints, and classical dynamics. If the symplectic form has null directions, stop and resolve the constraints or gauge directions before imposing naïve CCR.
2. Choose elementary observables. Select a real, adjoint-compatible generating space on which the brackets are under control. Do not silently extend the prescription to all nonlinear functions.
3. Define the quantum algebra. Replace the brackets of bosonic linear generators by CCR, or the graded brackets of fermionic generators by CAR. State whether a polynomial or exponentiated presentation is being used.
4. Check algebraic consistency. Verify adjoints, antisymmetry or grading, the Jacobi identity, central terms, and preservation of the intended symmetries.
5. Choose a state or representation. A state-first route can use GNS; a representation-first route must still identify the state used for predictions. For unbounded generators, specify a common domain.
6. Represent the dynamics. Check that time evolution acts consistently on the algebra and, when a Hamiltonian is used, that the proposed representation supports it.
7. Compare two descriptions. Reconstruct the smeared field relations from mode commutators, or project the field relations onto a complete mode basis. Agreement tests normalization and double counting independently of the initial presentation.

Steps 1–4 output a quantum algebra $\mathcal A$. A predictive model adds a representation $\rho$, a state $\omega$, and dynamics. For finitely many unconstrained linear modes, the algebraic steps are short. The main difficulty moves to composite-operator ordering, domains, state selection, constraints, and the limit of infinitely many modes. The procedure should stop rather than claim a complete quantum theory when one of those ingredients is essential but unresolved.

Worked application: free scalar normal modes

Regulate the free real scalar field by placing it in a finite spatial box with boundary conditions that make $-\nabla^2+m^2$ self-adjoint and remove the boundary term in integration by parts. Retain a finite set of independent real orthonormal eigenmodes satisfying

$$(-\nabla^2+m^2)u_\alpha =\omega_\alpha^2u_\alpha, \qquad \int_V \mathrm d^3\mathbf x\, u_\alpha(\mathbf x)u_\beta(\mathbf x) =\delta_{\alpha\beta}.$$

Take $m>0$, or treat the massless zero mode separately, so that every retained frequency is positive. At the reference time $t_0$,

$$\begin{aligned} \phi(t_0,\mathbf x)&= \sum_{\alpha=1}^{N}q_\alpha u_\alpha(\mathbf x),\\ \pi(t_0,\mathbf x)&= \sum_{\alpha=1}^{N}p_\alpha u_\alpha(\mathbf x), \end{aligned}$$

Using a real mode basis avoids the constraint and double counting that arise if the coefficients of complex Fourier modes at $\mathbf k$ and $-\mathbf k$ are treated as independent. The classical Hamiltonian and brackets are

$$H_N^{\mathrm{cl}}=\frac12\sum_{\alpha=1}^{N} \left(p_\alpha^2+\omega_\alpha^2q_\alpha^2\right), \qquad \{q_\alpha,p_\beta\}=\delta_{\alpha\beta}.$$

The preceding Klein–Gordon modes page derives the classical mode normalization and positive-frequency split. Here the canonical step is to introduce abstract generators satisfying $Q_\alpha^*=Q_\alpha$ and $P_\alpha^*=P_\alpha$, with

$$[Q_\alpha,P_\beta] =i\delta_{\alpha\beta}\mathbf 1, \qquad [Q_\alpha,Q_\beta]=[P_\alpha,P_\beta]=0.$$

Define

$$a_\alpha=c_\alpha Q_\alpha+i d_\alpha P_\alpha, \qquad a_\alpha^*=c_\alpha Q_\alpha-i d_\alpha P_\alpha,$$

where

$$c_\alpha=\sqrt{\frac{\omega_\alpha}{2}}, \qquad d_\alpha=\frac{1}{\sqrt{2\omega_\alpha}}.$$

The oscillator algebra follows directly:

$$\begin{aligned} [a_\alpha,a_\beta^*] &=(c_\alpha d_\beta+d_\alpha c_\beta) \delta_{\alpha\beta}\mathbf 1\\ &=\delta_{\alpha\beta}\mathbf 1, \end{aligned} \qquad [a_\alpha,a_\beta]=0.$$

For the finite system the Hamiltonian element can then be written algebraically as

$$\mathsf H_N :=\frac12\sum_{\alpha=1}^{N} \left(P_\alpha^2+\omega_\alpha^2Q_\alpha^2\right) =\sum_{\alpha=1}^{N} \omega_\alpha \left(a_\alpha^*a_\alpha+\frac12\mathbf 1\right).$$

This relation still does not choose a vacuum. In the usual Fock representation one adds a normalized vector $\Omega_0$ satisfying $\rho(a_\alpha)\Omega_0=0$ for every $\alpha$. The corresponding state and its particle interpretation depend on that positive-frequency choice; neither follows from the bare $Q$–$P$ commutators.

Independent field–mode check

Project test functions onto the retained modes,

$$f_\alpha=\int_V \mathrm d^3\mathbf x\, u_\alpha(\mathbf x)f(\mathbf x), \qquad g_\alpha=\int_V \mathrm d^3\mathbf x\, u_\alpha(\mathbf x)g(\mathbf x),$$

and define

$$\Phi_N(f)=\sum_{\alpha=1}^{N}f_\alpha Q_\alpha, \qquad \Pi_N(g)=\sum_{\alpha=1}^{N}g_\alpha P_\alpha.$$

The mode algebra gives

$$[\Phi_N(f),\Pi_N(g)] =i\sum_{\alpha=1}^{N}f_\alpha g_\alpha\mathbf 1.$$

Equivalently, the unsmeared kernel is

$$K_N(\mathbf x,\mathbf y)= \sum_{\alpha=1}^{N} u_\alpha(\mathbf x)u_\alpha(\mathbf y),$$

the integral kernel of the projection onto the retained modes. It is not the full delta distribution at finite $N$. As a complete orthonormal basis is restored, $K_N$ converges distributionally to the box delta and the sum converges to $\int_V \mathrm d^3\mathbf x\,f(\mathbf x)g(\mathbf x)$. This comparison between field-first smearing and mode-first quantization detects a missing normalization factor, an incomplete basis, or double counting before the regulator is removed Schwartz 2014, § 2.3, pp. 20–26.

Where the prescription stops

Canonical quantization has produced a consistent finite-mode algebra and a check of its normalization. It has not solved every problem needed for a field theory:

• Ordering: nonlinear classical observables require an additional prescription. Deformation quantization and star products provide a systematic alternative viewpoint.
• Constraints: gauge constraints or second-class constraints must be reduced or treated with an appropriate constrained-quantization method before naïve CCR can be trusted.
• Domains: formal commutator identities do not establish self-adjointness or equality on arbitrary Hilbert-space vectors.
• Infinite systems: the limit $N\to\infty$ can change which representations exist or are unitarily equivalent, so finite-degree-of-freedom intuition is not a uniqueness theorem for QFT.
• State selection: dynamics, symmetries, boundary conditions, or physical preparation must supply criteria beyond the algebraic relations.

A practical stop rule is: do not call the quantization complete until the chosen generators form an adjoint-compatible algebra, the proposed state is positive, the representation has adequate domains, the dynamics preserves the relations, and an independent normalization check agrees. Failure of any item identifies the next mathematical problem rather than an ignorable technicality. The quantization of the real scalar field carries these ingredients into the full free-field construction.

Check your understanding

1. Retrieval. State in one sentence each what the algebra $\mathcal A$, representation $\rho$, and state $\omega$ provide.
2. Distinction. Explain why $[Q,P]=i\mathbf 1$ does not select a vacuum, even when a Fock representation is available.
3. Normalization. Starting from the displayed $c_\alpha,d_\alpha$, recover $[a_\alpha,a_\beta^*]=\delta_{\alpha\beta}\mathbf 1$ and identify the field-space kernel obtained with only $N$ modes.
4. Failure mode. A calculation treats the complex Fourier coefficients at $\mathbf k$ and $-\mathbf k$ as independent for a real field and obtains an exact delta distribution at finite $N$. Identify both errors.
5. Transfer. For finitely many coupled oscillators with $[Q_j,P_k]=i\delta_{jk}\mathbf 1$, let $\widetilde Q_\alpha=\sum_jO_{\alpha j}Q_j$ and $\widetilde P_\alpha=\sum_jO_{\alpha j}P_j$ for a real orthogonal matrix $O$. Check the transformed CCR.
6. Handoff. Decide where to continue for (a) the complete free-scalar operator construction, (b) self-adjointness and common domains, and (c) fermionic canonical relations.

Answer and repair routes

1. $\mathcal A$ specifies generators and relations; $\rho$ realizes them as operators on a Hilbert space and common domain; $\omega$ assigns positive normalized expectation values. Revisit Algebra, representation, and state if any two answers coincide.
2. The CCR contain neither a positive-frequency split nor a positive functional. A vacuum requires additional dynamics and a state condition such as $\rho(a_\alpha)\Omega_0=0$; compare vacua, states, and representations.
3. Since $2c_\alpha d_\alpha=1$, the canonical commutator gives $[a_\alpha,a_\beta^*]=\delta_{\alpha\beta}\mathbf 1$. The finite field kernel is $K_N(\mathbf x,\mathbf y)=\sum_{\alpha=1}^Nu_\alpha(\mathbf x)u_\alpha(\mathbf y)$, a projection rather than the full delta distribution.
4. Reality relates the coefficients at opposite momenta, so counting both independently duplicates degrees of freedom. A finite basis yields $K_N$, not an exact delta; repair the calculation with independent real modes and the field–mode check.
5. Orthogonality gives $[\widetilde Q_\alpha,\widetilde P_\beta]=i\sum_jO_{\alpha j}O_{\beta j}\mathbf 1=i\delta_{\alpha\beta}\mathbf 1$, with the $\widetilde Q$–$\widetilde Q$ and $\widetilde P$–$\widetilde P$ commutators still zero. Thus a real canonical normal-mode rotation preserves the CCR.
6. Continue to quantizing the real scalar field for (a), unbounded operators, domains, closure, and adjoints for (b), and canonical quantization of the free Dirac field for (c).

Common pitfalls

Treating the delta distribution as a number. The point-labelled equal-time commutator is shorthand. Test it only after smearing; with a finite mode cutoff, expect the projected kernel $K_N$, not an exact delta.

Calling the algebra a Hilbert space. CCR define abstract relations. Spectra, domains, vacuum vectors, and particle-number operators acquire their operator meaning only in a representation.

Claiming that the CCR select the vacuum. The same canonical algebra can support many states. A vacuum condition uses dynamics and a positive-frequency choice in addition to the commutators.

Quantizing every classical product mechanically. Once generators fail to commute, classical identities such as $qp=pq$ no longer choose a unique ordering. State the composite-observable prescription or stop at the linear algebra.

Using bosonic commutators for fermionic variables. Fermionic generators require CAR and graded signs. This change cannot be repaired by altering a normalization after the calculation.

References

• Hollands, Stefan, and Robert M. Wald. “Quantum Fields in Curved Spacetime.” Physics Reports 574 (2015): 1–35. DOI. Open PDF.
• Schwartz, Matthew D. Quantum Field Theory and the Standard Model. Cambridge: Cambridge University Press, 2014. DOI.