Canonical Commutation Relations

For one canonical coordinate $X$ and its conjugate momentum $P$, the canonical commutation relation is

$$[X,P]=i\hbar I.$$

For $n$ Cartesian degrees of freedom, the canonical algebra is

$$[X_i,P_j]=i\hbar\delta_{ij}I,$$ $$[X_i,X_j]=0, \qquad [P_i,P_j]=0.$$

These are operator relations, not numerical measurement outcomes. They state that each $P_i$ is conjugate to $X_i$, while distinct Cartesian canonical pairs commute in the standard unconstrained theory.

The algebra has several inseparable meanings:

• $X$ and $P$ are incompatible sharp observables.
• $P$ generates translations of $X$.
• $X$ generates translations of $P$.
• position- and momentum-space wavefunctions are Fourier transforms.
• the uncertainty bound $\Delta X\,\Delta P\ge\hbar/2$ follows.
• commutators with functions of $X$ and $P$ reproduce the canonical structure of Hamiltonian dynamics.

The compact equation hides important qualifications. $X$ and $P$ are unbounded, so the raw commutator holds on a suitable common invariant domain, not as an everywhere-defined bounded-operator identity. Its exponentiated Weyl form is often the cleaner rigorous statement.

For the precise relation between bounded unitary shifts and unbounded generators, see Weyl Form of the Canonical Commutation Relations.

Helpful background. Commutators supplies the product rule; wavefunctions as representations explains the position and momentum descriptions.

Why the Pair Is Called Canonical

In classical Hamiltonian mechanics, canonical phase-space coordinates satisfy

$$\{q_i,p_j\}_{\mathrm{PB}}=\delta_{ij},$$ $$\{q_i,q_j\}_{\mathrm{PB}}=0, \qquad \{p_i,p_j\}_{\mathrm{PB}}=0.$$

Canonical quantization replaces these basic brackets schematically by

$$\{\,\cdot\,,\,\cdot\,\}_{\mathrm{PB}} \longrightarrow \frac1{i\hbar} [\,\cdot\,,\,\cdot\,].$$

Applying this rule to the basic coordinates gives the canonical commutation relations. The correspondence is structurally valuable but is not a universal quantization algorithm: nonlinear observables introduce ordering ambiguities, and no prescription preserves every classical Poisson bracket exactly.

The classical geometry belongs to Poisson Brackets, while the distinction between constructing a quantum model and taking a classical limit is developed in Quantization vs Classical Limit.

Schrödinger Representation on the Line

Take

$$\mathcal H=L^2(\mathbb R,dx).$$

For the calculation, take $\psi$ in Schwartz space: smooth functions whose derivatives decay faster than any inverse power of $|x|$. This space is preserved by both operators. The self-adjoint operators extend these actions to the domains specified below. Position multiplies and momentum differentiates:

$$\begin{aligned} (X\psi)(x) &=x\psi(x), \\ (P\psi)(x) &=-i\hbar\frac{d\psi}{dx}. \end{aligned}$$

The two orderings act as

$$(XP\psi)(x) = -i\hbar x\psi'(x),$$

and

$$\begin{aligned} (PX\psi)(x) &=-i\hbar \frac{d}{dx} \left(x\psi(x)\right) \\ &=-i\hbar\psi(x) -i\hbar x\psi'(x). \end{aligned}$$

Subtracting gives

$$([X,P]\psi)(x) = i\hbar\psi(x).$$

Thus

$$[X,P]=i\hbar I$$

on the chosen common domain. The identity operator is often suppressed in informal notation, but it matters when comparing operator types, taking traces, or generalizing the algebra.

Momentum Representation

Choose the Fourier convention

$$\widetilde\psi(p) = \frac1{\sqrt{2\pi\hbar}} \int_{-\infty}^{\infty} e^{-ipx/\hbar} \psi(x)\,dx.$$

For Schwartz functions the Fourier integral is ordinary; for a general $L^2$ wavefunction it is defined by the unitary $L^2$ extension. In momentum space,

$$\begin{aligned} (P\widetilde\psi)(p) &=p\widetilde\psi(p), \\ (X\widetilde\psi)(p) &=i\hbar\frac{d\widetilde\psi}{dp}. \end{aligned}$$

Direct calculation again yields

$$[X,P]\widetilde\psi = i\hbar\widetilde\psi.$$

The sign of the derivative representation tracks the sign chosen in the Fourier kernel. A different convention is equally valid if every transform, inverse transform, and translation phase is changed consistently. See Fourier-Transform Conventions and Position and Momentum Representations.

Several Degrees of Freedom

On

$$\mathcal H=L^2(\mathbb R^n,d^nx),$$

the Cartesian Schrödinger representation is

$$\begin{aligned} (X_i\psi)(\mathbf x) &=x_i\psi(\mathbf x), \\ (P_j\psi)(\mathbf x) &=-i\hbar \frac{\partial\psi}{\partial x_j}. \end{aligned}$$

The product rule gives

$$([X_i,P_j]\psi)(\mathbf x) = i\hbar\delta_{ij}\psi(\mathbf x).$$

Thus

$$[X_i,P_j]=i\hbar\delta_{ij}I.$$

When $i\ne j$, differentiating with respect to $x_j$ does not act on the factor $x_i$, so the commutator vanishes.

For many distinguishable particles, add a particle label $\alpha$:

$$[X_{\alpha i},P_{\beta j}] = i\hbar \delta_{\alpha\beta} \delta_{ij}I.$$

Canonical variables belonging to different particle labels or different Cartesian directions commute in this representation.

Symplectic Matrix Form

Collect all canonical operators into

$$Z = (X_1,\ldots,X_n,P_1,\ldots,P_n)^{\mathsf T}.$$

Then the full algebra can be written compactly as

$$[Z_a,Z_b] = i\hbar\Omega_{ab}I,$$

where

$$\Omega = \begin{pmatrix} 0&I_n\\ -I_n&0 \end{pmatrix}.$$

The antisymmetric matrix $\Omega$ is the standard symplectic form in canonical coordinates. Real linear transformations $Z\mapsto SZ$ preserve the canonical commutation relations exactly when

$$S\Omega S^{\mathsf T}=\Omega.$$

This matrix form becomes especially useful for coupled oscillators, Gaussian states, and phase-space methods.

Commutators with Powers and Functions

The basic canonical relation and the commutator product rule imply

$$[X,P^n] = i\hbar nP^{n-1},$$

and

$$[P,X^n] = -i\hbar nX^{n-1}.$$

For differentiable functions, the corresponding derivative rules are

$$[X,f(P)] = i\hbar f'(P),$$

and

$$[P,g(X)] = -i\hbar g'(X).$$

The formulas are exact for polynomials on Schwartz space. More generally, the second identity follows from the ordinary product rule wherever $g\psi$ and its derivative lie in the relevant domains; the first follows by the same argument in momentum space. The spectral theorem defining $g(X)$ or $f(P)$ does not, by itself, justify a commutator derivative.

Momentum Generates Position Translations

Define the unitary translation operator

$$T(a) = \exp\left(-\frac{iaP}{\hbar}\right).$$

In the position representation,

$$(T(a)\psi)(x)=\psi(x-a).$$

Conjugating $X$ gives

$$T^{\dagger}(a)XT(a) = X+aI.$$

Direct substitution of the translated wavefunction proves this on $\mathcal D(X)$, which translations preserve. It agrees with the formal nested-commutator calculation: the first correction is

$$\frac{ia}{\hbar}[P,X]=aI,$$

and every higher nested commutator vanishes because $[P,I]=0$.

This is the operational generator meaning of the canonical relation: momentum is the infinitesimal generator of spatial translations. The symmetry-side development is Momentum Operator as Generator.

Position Generates Momentum Translations

Define the modulation or momentum-translation operator

$$M(b) = \exp\left(\frac{ibX}{\hbar}\right).$$

It acts in position space as

$$(M(b)\psi)(x) = e^{ibx/\hbar}\psi(x),$$

and shifts momentum under conjugation:

$$M^{\dagger}(b)PM(b) = P+bI.$$

The product rule verifies this on $\mathcal D(P)=H^1(\mathbb R)$, preserved by $M(b)$. Thus $X$ and $P$ generate translations along the two conjugate directions of phase space.

Weyl Form of the Canonical Relations

The two unitary shifts do not commute. Acting on a wavefunction shows

$$T(a)M(b) = e^{-iab/\hbar} M(b)T(a).$$

The mismatch is a phase in the center of the operator algebra. This is the Weyl relation, the exponentiated form of the canonical commutation relation.

For the real-line canonical pair, both orders shift the mean position by $a$ and mean momentum by $b$. Their final state vectors differ by $e^{-iab/\hbar}$. This is a global phase for either path alone; comparing the two paths coherently can turn it into a relative phase.

Unlike $X$ and $P$, the operators $T(a)$ and $M(b)$ are bounded and defined on the whole Hilbert space. Strongly continuous unitary families also recover self-adjoint generators through Stone’s theorem. For these reasons the Weyl form is often preferable in rigorous representation theory.

The group structure behind the phase is developed in Heisenberg Group.

Stone–von Neumann Uniqueness

For finitely many canonical degrees of freedom, every irreducible, strongly continuous representation of the Weyl relations with the same nonzero central parameter $\hbar$ is unitarily equivalent to the Schrödinger representation.

This result explains why position space and momentum space are two representations of one ordinary canonical quantum theory rather than distinct theories. The Fourier transform supplies the unitary map between them.

The hypotheses are essential:

• the number of degrees of freedom is finite;
• the representation is irreducible;
• the Weyl operators depend strongly continuously on their parameters;
• the central phase is fixed.

With infinitely many degrees of freedom, as in quantum field theory or thermodynamic limits, unitarily inequivalent representations can occur. The Stone–von Neumann Theorem explains the precise hypotheses and proof; its reference entry gives a compact statement.

No Exact Finite-Dimensional Canonical Pair

No finite matrices $X$ and $P$ can satisfy

$$[X,P]=i\hbar I.$$

For $d\times d$ matrices, cyclicity gives

$$\operatorname{Tr}[X,P]=0,$$

whereas

$$\operatorname{Tr}(i\hbar I_d) = i\hbar d\ne0.$$

This contradiction is one reason an exact canonical coordinate requires an infinite-dimensional Hilbert space. Finite truncations used in numerics can approximate low-energy matrix elements but must violate the canonical relation somewhere, often near the truncation boundary.

Position–Momentum Uncertainty

For a normalized state in the common product domain with finite variances, the Robertson inequality gives

$$\Delta X\,\Delta P \ge \frac12 \left| \langle[X,P]\rangle \right|.$$

Using the canonical relation,

$$\Delta X\,\Delta P \ge \frac{\hbar}{2}.$$

A quadratic-form derivation extends this real-line bound to all normalized states in $\mathcal D(X)\cap\mathcal D(P)$, without requiring both products to exist. The bound concerns standard deviations in one prepared state. It is not a claim about finite instrument resolution, and it does not say that either observable lacks a sharp eigenvalue in every generalized sense. The variance derivation, Gaussian equality states, and Fourier interpretation belong to Position–Momentum Uncertainty.

Recovering Hamiltonian Equations

For

$$H=\frac{P^2}{2m}+V(X),$$

with differentiable real $V$, the canonical derivative identities give the following formulas on a common domain where all products exist:

$$[X,H] = i\hbar\frac{P}{m},$$

and

$$[P,H] = -i\hbar V'(X).$$

The Heisenberg equations therefore read

$$\frac{dX_H}{dt} = \frac{P_H}{m},$$

and

$$\frac{dP_H}{dt} = -V'(X_H).$$

To read these as differential equations under a self-adjoint Hamiltonian, the evolving vectors must remain in the required domains and the derivatives must exist. These are operator versions of the classical canonical equations. They do not make the full quantum dynamics classical: $X_H$ and $P_H$ remain operators, states can spread and interfere, and nonlinear expectation values need not close on the expectations alone. The dynamics-side treatment is Commutator Dynamics.

Canonical and Kinetic Momentum in a Magnetic Field

For a particle of charge $q$ with minimal coupling in a sufficiently smooth vector potential $\mathbf A(\mathbf X)$, the canonical momentum in the Cartesian Schrödinger representation is

$$\mathbf P=-i\hbar\boldsymbol\nabla.$$

The kinetic, or mechanical, momentum is

$$\boldsymbol\Pi = \mathbf P-q\mathbf A(\mathbf X).$$

Both satisfy

$$[X_i,P_j] = [X_i,\Pi_j] = i\hbar\delta_{ij}I,$$

because functions of position commute with position. Their component commutators differ:

$$[P_i,P_j]=0,$$

but

$$[\Pi_i,\Pi_j] = iq\hbar \sum_k\epsilon_{ijk}B_k(\mathbf X).$$

The kinetic momentum is proportional to velocity and is gauge covariant; canonical momentum is the variable conjugate to position and the generator appearing in the canonical algebra. Conflating the two hides the magnetic field in precisely the commutator where it has physical consequences.

Domains Are Part of the Relation

On $L^2(\mathbb R)$, a natural position-operator domain is

$$\mathcal D(X) = \left\lbrace \psi\in L^2(\mathbb R): x\psi\in L^2(\mathbb R) \right\rbrace.$$

The self-adjoint real-line momentum has domain $H^1(\mathbb R)$: square-integrable functions with square-integrable weak derivative. The Schwartz space

$$\mathcal S(\mathbb R)$$

is a convenient dense common invariant core: multiplication by $x$ and differentiation both preserve it, and the calculation

$$[X,P]\psi=i\hbar\psi$$

is valid there.

The raw identity does not mean that $XP$, $PX$, and $I$ are bounded operators defined on every vector in $L^2(\mathbb R)$. Boundary conditions also matter. On a finite interval or compact configuration space, self-adjoint momentum operators depend on boundary data, and global canonical coordinates may require a different formulation.

The general operator-domain framework is developed in Unbounded Operators.

Situations Requiring Modification

Periodic coordinates

An angular coordinate is periodic, while angular momentum has a discrete spectrum. A globally defined self-adjoint angle operator cannot satisfy the naive relation

$$[\Phi,L_z]=i\hbar I$$

on all angular-momentum eigenstates. Unitary phase operators such as $e^{i\Phi}$ and their Weyl-type relations are safer global objects.

Energy and time

Ordinary nonrelativistic quantum mechanics treats time as an external parameter, not automatically as a self-adjoint operator $T$ satisfying $[T,H]=i\hbar I$. Energy–time uncertainty therefore requires a different analysis from the position–momentum pair; see Energy–Time Uncertainty.

Constraints and curved coordinates

Constrained systems may require reduced phase-space variables or Dirac brackets before quantization. In curvilinear coordinates, the inner-product measure and self-adjointness can add terms to naive differential momentum operators. The Cartesian formula $P_j=-i\hbar\partial_j$ should not be copied unchanged into every coordinate system.

Quantum fields

For a canonical field and its conjugate momentum, the discrete index becomes a spatial point, schematically giving an equal-time relation

$$[\Phi(t,\mathbf x),\Pi(t,\mathbf y)] = i\hbar\delta^{(d)}(\mathbf x-\mathbf y)I.$$

This is an operator-valued-distribution statement, not an ordinary operator identity at one point. The field-theory construction and constraint caveats belong to From Phase Space to Canonical Quantization.

Sign and Convention Checks

With

$$P=-i\hbar\frac{d}{dx},$$

the canonical sign is

$$[X,P]=+i\hbar I, \qquad [P,X]=-i\hbar I.$$

With

$$T(a)=e^{-iaP/\hbar}, \qquad M(b)=e^{ibX/\hbar},$$

the Weyl phase is

$$T(a)M(b) = e^{-iab/\hbar}M(b)T(a).$$

Changing the Fourier kernel or choosing the opposite active/passive translation convention changes several signs together. A reliable check is to act on a test wavefunction and verify both the shift direction and the commutator.

Common Pitfalls

A formal commutator need not imply the Weyl relation. The Schrödinger translations above verify the bounded identity directly. For arbitrary unbounded operators, additional analytic assumptions are needed to pass from an infinitesimal algebra to finite unitary shifts.

Canonical and kinetic momentum differ. In a magnetic field, the position commutators agree but the kinetic momentum components need not commute. The magnetic field appears in their commutator.

Boundary conditions change the problem. Real-line Cartesian formulas do not automatically apply to angles, finite intervals, constrained coordinates, or fields evaluated at a point. Check the configuration space and operator domains first.

Exercises

Exercise 1: Position-space verification

For a Schwartz function $\psi$, use

$$\begin{aligned} (X\psi)(x) &=x\psi(x), \\ (P\psi)(x) &=-i\hbar\psi'(x). \end{aligned}$$

to verify $[X,P]\psi=i\hbar\psi$.

Solution

The two products are

$$(XP\psi)(x) = -i\hbar x\psi'(x),$$

and

$$(PX\psi)(x) = -i\hbar \left( \psi(x)+x\psi'(x) \right).$$

Subtracting,

$$\begin{aligned} ([X,P]\psi)(x) &=(XP\psi)(x) -(PX\psi)(x) \\ &=i\hbar\psi(x). \end{aligned}$$

Because $\psi\in\mathcal S(\mathbb R)$, multiplication and differentiation remain inside a common invariant domain throughout the calculation.

Exercise 2: Momentum-space verification

For a Schwartz wavefunction in momentum space, let

$$\begin{aligned} (P\widetilde\psi)(p) &=p\widetilde\psi(p), \\ (X\widetilde\psi)(p) &=i\hbar\widetilde\psi'(p). \end{aligned}$$

Verify the same canonical commutator.

Solution

Compute

$$\begin{aligned} (XP\widetilde\psi)(p) &= i\hbar\frac{d}{dp} \left(p\widetilde\psi(p)\right) \\ &=i\hbar\widetilde\psi(p) +i\hbar p\widetilde\psi'(p). \end{aligned}$$

whereas

$$(PX\widetilde\psi)(p) = i\hbar p\widetilde\psi'(p).$$

Their difference is

$$[X,P]\widetilde\psi = i\hbar\widetilde\psi.$$

Exercise 3: Functions of canonical operators

Work on Schwartz space. Use the product rule and $[X,P]=i\hbar I$ to show

$$[X,P^n]=i\hbar nP^{n-1}.$$

Then find $[X,P^2/(2m)]$.

Solution

The general power identity is

$$[X,P^n] = \sum_{r=0}^{n-1} P^r[X,P]P^{n-1-r}.$$

Since $[X,P]=i\hbar I$ is central, every summand equals $i\hbar P^{n-1}$. There are $n$ terms, so

$$[X,P^n] = i\hbar nP^{n-1}.$$

For the kinetic energy,

$$\left[X,\frac{P^2}{2m}\right] = \frac{i\hbar}{m}P.$$

Exercise 4: Weyl phase from wavefunctions

Let

$$(T(a)\psi)(x)=\psi(x-a),$$

and

$$(M(b)\psi)(x)=e^{ibx/\hbar}\psi(x).$$

Show that

$$T(a)M(b) = e^{-iab/\hbar}M(b)T(a).$$

Solution

Applying $M(b)$ first and then translating gives

$$\begin{aligned} (T(a)M(b)\psi)(x) &=(M(b)\psi)(x-a) \\ &=e^{ib(x-a)/\hbar}\psi(x-a) \\ &=e^{-iab/\hbar} \times(M(b)T(a)\psi)(x). \end{aligned}$$

This proves the operator relation.

Exercise 5: Hamiltonian equations

For

$$H=\frac{P^2}{2m}+V(X),$$

take differentiable $V$ and a common domain on which the written products exist. Derive the two commutators and hence the Heisenberg equations wherever the time derivatives are defined.

Solution

Because $[X,V(X)]=0$,

$$[X,H] = \frac1{2m}[X,P^2] = \frac{i\hbar}{m}P.$$

Also $[P,P^2]=0$, while the canonical derivative rule gives

$$[P,V(X)] = -i\hbar V'(X).$$

Using $\dot A_H=(i/\hbar)[H_H,A_H]$,

$$\dot X_H=\frac{P_H}{m}, \qquad \dot P_H=-V'(X_H).$$

Exercise 6: Kinetic momentum in a magnetic field

Let

$$\Pi_i=P_i-qA_i(\mathbf X),$$

in the Cartesian Schrödinger representation with smooth $A_i$, acting on compactly supported smooth test functions. Show that

$$[\Pi_i,\Pi_j] = iq\hbar \sum_k\epsilon_{ijk}B_k(\mathbf X).$$

Solution

Functions of position commute with one another, and

$$[P_i,f(\mathbf X)] = -i\hbar\partial_i f(\mathbf X).$$

Therefore

$$\begin{aligned} [\Pi_i,\Pi_j] &=-q[P_i,A_j]-q[A_i,P_j] \\ &=iq\hbar \left( \partial_iA_j-\partial_jA_i \right). \end{aligned}$$

Using

$$\partial_iA_j-\partial_jA_i = \sum_k\epsilon_{ijk}B_k$$

gives the result.

Exercise 7: Finite-dimensional obstruction

Prove that no $d\times d$ matrices with $d>0$ can satisfy

$$[X,P]=i\hbar I_d.$$

Solution

Cyclicity of the finite-dimensional trace gives

$$\operatorname{Tr}[X,P] = \operatorname{Tr}(XP)-\operatorname{Tr}(PX) =0.$$

The proposed right-hand side has trace

$$\operatorname{Tr}(i\hbar I_d) = i\hbar d,$$

which is nonzero. The two sides cannot be equal.

Exercise 8: Why angle is different

Suppose a normalized angular-momentum eigenstate satisfies

$$L_z|m\rangle=m\hbar|m\rangle$$

and lies in the domains of both $\Phi L_z$ and $L_z\Phi$. Show that the naive identity $[\Phi,L_z]=i\hbar I$ cannot hold on $|m\rangle$.

Solution

Take the diagonal matrix element. The left-hand side is

$$\begin{aligned} \langle m|[\Phi,L_z]|m\rangle &=m\hbar\langle m|\Phi|m\rangle -m\hbar\langle m|\Phi|m\rangle \\ &=0. \end{aligned}$$

The proposed right-hand side gives

$$\langle m|i\hbar I|m\rangle=i\hbar,$$

a contradiction. The assumptions behind a global self-adjoint angle operator and the naive canonical domain cannot all hold simultaneously. Periodicity is better encoded with unitary phase operators and Weyl-type relations.

References

• G. B. Folland, Harmonic Analysis in Phase Space, Princeton University Press, 1989.
• B. C. Hall, Quantum Theory for Mathematicians, Springer, 2013. Publisher.
• J. J. Sakurai and J. Napolitano, Modern Quantum Mechanics, 3rd ed., Cambridge University Press, 2020.