Fundamental Poisson Brackets

Page one of every quantum mechanics textbook displays the commutator $[\hat q,\hat p]=i\hbar$ as if it were quantum theory's founding axiom. It is actually a quotation — of a statement classical mechanics had been making for a century: $\{q,p\}=1$ . Together with its silent partners $\{q,q\}=0$ and $\{p,p\}=0$ , this is the set of fundamental Poisson brackets, and this page is their biography. Their power is wildly out of proportion to their size. The bracket machinery of Poisson Brackets: Definition assigns a number to every pair of observables — infinitely many brackets, one for each pair of functions on phase space. The fundamental brackets are the claim that three lines of that infinite table determine all the rest: bilinearity and the Leibniz rule propagate $\{q_i,p_j\}=\delta_{ij}$ outward until every bracket of every pair of observables is fixed. Three atoms, then chemistry. That makes the table a fingerprint. Coordinates that satisfy it are canonical — Hamilton's equations, conservation theorems, the whole formalism runs in them unchanged; coordinates that fail it, even by a constant factor, are impostors. Checking the fingerprint against a candidate change of variables is precisely the symplectic condition $MJM^{\mathsf T}=J$ in component form, a fact we prove in both directions. And when Dirac needed to know what classical structure the strange new quantum products were shadowing, it was this table — with $1$ traded for $i\hbar$ — that answered. The operational art of wielding the fingerprint as a yes/no test is the business of Poisson Brackets as a Test for Canonicity; here we establish what the fingerprint is and why it determines everything.

Key takeaways

The fundamental Poisson brackets $\{q_i,q_j\}=0$ , $\{p_i,p_j\}=0$ , $\{q_i,p_j\}=\delta_{ij}$ are the algebraic fingerprint of a canonical chart — compactly, $\{\eta_a,\eta_b\}=J_{ab}$ .
They generate everything: bilinearity plus the Leibniz rule reduce the bracket of any two observables to the table, so knowing three lines means knowing every bracket.
A transformation preserves the fundamental brackets iff its Jacobian is symplectic, $MJM^{\mathsf T}=J$ — the fingerprint and the symplectic condition are one statement in two notations.
The table is chart-specific, not automatic: $(2q,\,p/2)$ keeps it, $(q,\,2p)$ breaks it ( $\{Q,P\}=2$ ), and the "obvious" pair $(x,v)$ fails by a factor $1/m$ .
Point transformations $Q=Q(q)$ with induced momenta $P=(\partial q/\partial Q)\,p$ pass automatically — which is why Lagrangian mechanics never noticed the issue.
Under Dirac's correspondence the unit bracket becomes the canonical commutator: $\{q_i,p_j\}=\delta_{ij}\to[\hat q_i,\hat p_j]=i\hbar\,\delta_{ij}$ .
Key idea: in any canonical chart the only nonzero fundamental brackets are $\{q_i,p_j\}=\delta_{ij}$ ; this single table, propagated by bilinearity and Leibniz, fixes every Poisson bracket and is equivalent to the symplectic condition.

The core insight

In any canonical chart the only nonzero fundamental brackets are $\{q_i,p_j\}=\delta_{ij}$ ; this single fact, propagated by bilinearity and Leibniz, fixes every Poisson bracket and is equivalent to the symplectic condition.

Where this lives today

The table's modern career is largely a verification career: whenever physicists build new coordinates or new discretizations, "check the fundamental brackets" is the unit test. Structure-preserving ("symplectic") integrators — the workhorses of solar-system ephemerides, particle-accelerator tracking, and molecular dynamics — are precisely integration schemes whose one-step map has a Jacobian obeying $MJM^{\mathsf T}=J$ ; the code review for a new integrator literally includes verifying the discrete map's fundamental brackets, because a scheme that drifts the table by even $10^{-6}$ per step manufactures spurious energy over the $10^9$ -step hauls of a Milky Way or beam-lifetime simulation. The quantum side of the fingerprint is equally operational. A Bogoliubov transformation — the mixing of creation and annihilation operators used to describe superfluids, squeezed light, and Hawking radiation — is defined as a linear map preserving the canonical commutators, i.e. a symplectic matrix acting on the quantum table; gravitational-wave detectors now inject squeezed vacuum states engineered by exactly such transformations, and the squeezing hardware's calibration is a live check that $[\hat q,\hat p]=i\hbar$ survives the optical processing. The century-old three-line table is, daily, both a compiler check for planetary codes and a hardware spec for quantum optics.

The physical picture

Phase space is not a bare set of $2n$ numbers — it is stitched. Each coordinate $q_i$ is sewn to exactly one partner $p_i$ , with unit thread tension; everything else is unstitched. The fundamental brackets are the stitching pattern written as algebra: positions ignore positions, momenta ignore momenta, and each $q_i$ responds — at unit rate, in the flow-meter sense of Poisson Brackets: Definition — only to the flow generated by its own conjugate $p_i$ . Momentum generates translation of its own coordinate; every other pairing is silent.

Figure

The canonical fingerprint — a chart earns the name canonical by matching the three-line table, and the table is the symplectic matrix J read entry by entry

The second picture is generative. Suppose you know only the three lines of the table, plus the grammar of the bracket — bilinearity and the Leibniz product rule from Algebra of Poisson Brackets. Then you can compute $\{q^2,\,qp\}$ without touching a partial derivative: Leibniz peels the products apart until only brackets of bare coordinates remain, and those are table lookups. The same peeling works for any polynomial, hence (taking limits) for any smooth observable. The infinite bracket table of phase space is data-compressed into three lines; everything else is grammar.

Figure

Three atoms generate all brackets — Leibniz and bilinearity peel any bracket down to table lookups, so preserving the atoms preserves the universe

The third picture explains why the table matters beyond bookkeeping: it is the certificate of citizenship for coordinates. Hamilton's equations, the evolution law $\dot f=\{f,H\}$ , Liouville volume conservation — all of it is built on the symplectic matrix $J$ , and the table says exactly "my coordinates read out $J$ ." Change variables and the table becomes a nontrivial claim: polar coordinates with their induced momenta still match it; the innocent-looking pair (position, velocity) does not. When the fingerprint survives a change of chart, all of Hamiltonian mechanics transfers verbatim; when it does not, formulas silently break by exactly the factor the table reports. That factor-checking is a craft of its own, developed in Poisson Brackets as a Test for Canonicity.

The table, and its matrix form

Recall from Poisson Brackets: Definition — one line, no re-derivation — that $\{f,g\}=\sum_i(\partial_{q_i}f\,\partial_{p_i}g-\partial_{p_i}f\,\partial_{q_i}g)$ , and that evaluating this on the coordinate functions themselves collapses the sum by Kronecker deltas. The result is the table this page is about:

\{q_i,q_j\}=0,\qquad \{p_i,p_j\}=0,\qquad \{q_i,p_j\}=\delta_{ij} \qquad (1)

where: $q_i,p_i$ — canonical pairs, $i=1,\dots,n$ ; $\delta_{ij}$ — Kronecker delta, $1$ if $i=j$ , else $0$ . Read it physically: the flow generated by $p_j$ (a rigid translation of $q_j$ ) moves $q_i$ at rate $\delta_{ij}$ — unit rate for its own partner, not at all for anyone else's — and moves no momentum. Same-type variables are mutually invisible. Stack the coordinates into a single $2n$ -column and the three statements fuse into one. Define

\eta=(q_1,\dots,q_n,\,p_1,\dots,p_n)^{\mathsf T},\qquad J=\begin{pmatrix}0&I_n\\-I_n&0\end{pmatrix} \qquad (2)

where: $\eta$ — the phase-space point as a column; $J$ — the symplectic matrix, satisfying $J^{\mathsf T}=-J$ , $J^2=-I_{2n}$ , $J^{-1}=-J$ . Then the entire table is the single equation

\{\eta_a,\eta_b\}=J_{ab},\qquad a,b=1,\dots,2n \qquad (3)

where: the top-right block $I_n$ of $J$ holds the $\{q_i,p_j\}=\delta_{ij}$ entries, the bottom-left $-I_n$ their antisymmetric mirrors $\{p_i,q_j\}=-\delta_{ij}$ , and the zero blocks the vanishing same-type brackets. This is worth staring at: the coordinate functions, fed into the bracket, read out the symplectic matrix entry by entry. The geometry of phase space and the algebra of the table are not analogous — they are the same object in two fonts. In the language of Symplectic Group and Phase-Space Geometry, equation (3) says the canonical chart presents the symplectic form in its normal form. One consequence follows immediately and sets up everything below. For any smooth $f,g$ the bracket can be written through $\eta$ -gradients as $\{f,g\}=(\partial_\eta f)^{\mathsf T}J\,(\partial_\eta g)$ — and since $J_{ab}=\{\eta_a,\eta_b\}$ , this reads

\{f,g\}=\sum_{a,b=1}^{2n}\frac{\partial f}{\partial\eta_a}\,\frac{\partial g}{\partial\eta_b}\,\{\eta_a,\eta_b\} \qquad (4)

where: the double sum runs over all $2n$ phase-space directions. Equation (4) is the reduction principle in embryo: the bracket of anything is a weighted sum of fundamental brackets, with weights that are ordinary partial derivatives. The table is the only bracket data the formalism ever consumes.

Derivation AEvery bracket reduces to the fundamental table

We are proving the compression claim: the three lines of table (1), together with bilinearity and the Leibniz rule, determine the Poisson bracket of every pair of smooth observables — no further bracket data exists. The strategy has two prongs. First, a purely algebraic induction: Leibniz peels polynomial observables down to coordinate brackets, which are table lookups. Second, the analytic closure (4): for general smooth functions the chain-rule form shows the bracket consumes only $\{\eta_a,\eta_b\}$ . Watch how the two prongs meet — the algebra never needs the definition, only the grammar, which is why any structure satisfying the same grammar and table (quantum operators included) reproduces the same algebra.

Step 1. Base case: brackets of bare coordinates are the table itself, $\{\eta_a,\eta_b\}=J_{ab}$ . Constants bracket to zero with everything, $\{f,c\}=0$ , by bilinearity (bracket a constant multiple of $1$ and note $\{f,1\}=\{f,1\cdot 1\}=2\{f,1\}$ , forcing $\{f,1\}=0$ ). Why this trick is worth showing: it is the first taste of the method — no derivative was taken; the grammar alone forced the value.

Step 2. Inductive step on monomials: suppose brackets of $f$ with all monomials of degree $\le k$ are determined. A degree- $(k{+}1)$ monomial factors as $\eta_a m$ with $\deg m=k$ , and Leibniz gives

\{f,\eta_a m\}=\{f,\eta_a\}\,m+\eta_a\{f,m\}

Both right-hand brackets are determined (degree $1$ and degree $k$ ), so the left side is. By induction, $\{f,g\}$ is fixed for all polynomial $g$ ; running the same induction in the first slot (antisymmetry swaps slots at the cost of a sign) fixes all polynomial-polynomial brackets.

Step 3. Concretely, the peeling in the second diagram: $\{q^2,qp\}=q\{q^2,p\}+\{q^2,q\}p$ by Leibniz in the second slot; then $\{q^2,p\}=2q\{q,p\}=2q$ and $\{q^2,q\}=2q\{q,q\}=0$ by Leibniz in the first slot. Total: $\{q^2,qp\}=2q^2$ . Cross-check against the definition: $\partial_q(q^2)\,\partial_p(qp)-\partial_p(q^2)\,\partial_q(qp)=2q\cdot q-0=2q^2$ . Identical — the grammar route and the calculus route cannot disagree.

Step 4. Analytic closure for non-polynomial observables: expand any smooth $f,g$ in the chain-rule form. Because the bracket is a first-order derivation in each slot (Leibniz again, now for composite functions), the same peeling yields exactly equation (4): $\{f,g\}=\sum_{a,b}(\partial_{\eta_a}f)(\partial_{\eta_b}g)\{\eta_a,\eta_b\}$ . Every bracket is a linear combination of table entries with derivative coefficients; substituting $J_{ab}$ and writing out the blocks reproduces the coordinate definition of the bracket, confirming self-consistency. Why "preserve the table" then means "preserve everything": a change of variables under which the new coordinates satisfy the same table feeds identical atomic data into (4), so every bracket of every pair of observables comes out unchanged. This single sentence is the engine behind the canonicity test of the companion article. Verified: the peeled value $\{q^2,qp\}=2q^2$ matches the direct partial-derivative computation above; equation (4) with $\{\eta_a,\eta_b\}=J_{ab}$ written out in blocks reproduces the standard definition (Goldstein §9.5, Landau–Lifshitz §42).

The fingerprint theorem: table preserved iff Jacobian symplectic

Now the central structural fact. Take any invertible, smooth change of phase-space variables $\eta\to\zeta(\eta)$ — the candidate new chart $\zeta=(Q_1,\dots,Q_n,P_1,\dots,P_n)^{\mathsf T}$ — and ask: do the new coordinates still satisfy the table? The answer is controlled entirely by the transformation's Jacobian matrix,

M_{ab}=\frac{\partial\zeta_a}{\partial\eta_b} \qquad (5)

where: $M$ — a $2n\times2n$ matrix of partials, generally position-dependent. The chain rule converts the question into a matrix sandwich (Derivation B below):

\{\zeta_a,\zeta_b\}_{\eta}=\big(MJM^{\mathsf T}\big)_{ab} \qquad (6)

where: the subscript $\eta$ insists the brackets are computed in the old chart — the only chart currently certified. Equation (6) is a genuinely useful reading device: the full matrix of candidate fundamental brackets, all $4n^2$ of them, is one matrix product. Demanding the new chart carry the fingerprint, $\{\zeta_a,\zeta_b\}=J_{ab}$ , is therefore demanding

MJM^{\mathsf T}=J \qquad\Longleftrightarrow\qquad M^{\mathsf T}JM=J \qquad (7)

where: either form declares $M$ a symplectic matrix — the defining condition of the group $Sp(2n,\mathbb R)$ studied in Symplectic Group and Phase-Space Geometry, and the symplectic condition of Symplectic Condition. The fingerprint and the symplectic condition are one theorem in two scripts: brackets for calculators, matrices for geometers.

Go deeper · Derivation B · 2 minPreserving the fundamental brackets is exactly the symplectic condition

Derivation BPreserving the fundamental brackets is exactly the symplectic condition

We are proving the equivalence both ways: new variables $\zeta(\eta)$ satisfy the fundamental table iff the Jacobian $M$ obeys $MJM^{\mathsf T}=J$ , and this form is interchangeable with $M^{\mathsf T}JM=J$ . The route is the chain rule to establish identity (6), then a two-line matrix argument each way; the transpose equivalence needs one lemma, $M^{-1}=-JM^{\mathsf T}J$ , worth internalizing because it is the working formula for inverting symplectic matrices without Gaussian elimination. The pitfall to watch: (6) holds for any smooth invertible map — no canonicity assumed — so all the physics sits in whether the sandwich equals $J$ .

Step 1. Chain-rule identity. Each new coordinate $\zeta_a$ is a phase-space function, so equation (4) applies to the pair $(\zeta_a,\zeta_b)$ :

\{\zeta_a,\zeta_b\}_\eta=\sum_{c,d}\frac{\partial\zeta_a}{\partial\eta_c}\,\frac{\partial\zeta_b}{\partial\eta_d}\,\{\eta_c,\eta_d\}=\sum_{c,d}M_{ac}\,J_{cd}\,M_{bd}=\big(MJM^{\mathsf T}\big)_{ab}

The last equality reads the index pattern: $M_{bd}$ summed over its second index against $J$ 's second index is a transpose.

Step 2. Forward direction. If the new chart carries the fingerprint, $\{\zeta_a,\zeta_b\}_\eta=J_{ab}$ for all $a,b$ , then Step 1 says the matrices agree entry by entry: $MJM^{\mathsf T}=J$ . Nothing else is needed — the fingerprint is the symplectic condition, read componentwise.

Step 3. Reverse direction. If $MJM^{\mathsf T}=J$ , then Step 1 immediately returns $\{\zeta_a,\zeta_b\}_\eta=J_{ab}$ : the new variables satisfy the table. Both implications are trivial once (6) exists — the entire content of the theorem lives in the chain rule. Why the two directions are not vacuous: forward says "canonical charts are recognizable by their brackets" — a detection principle; reverse says "matching brackets certifies a canonical chart" — a construction license. The companion test article runs the license operationally.

Step 4. Transpose equivalence. From $MJM^{\mathsf T}=J$ , multiply on the left by $J^{-1}=-J$ : $-JMJM^{\mathsf T}=I$ , which exhibits the inverse explicitly, $M^{-1}=-JM^{\mathsf T}J$ — the inversion lemma. Now invert both sides of the sandwich itself: $(MJM^{\mathsf T})^{-1}=J^{-1}$ gives $(M^{\mathsf T})^{-1}J^{-1}M^{-1}=J^{-1}$ , and multiplying through by $-1$ trades every $J^{-1}=-J$ for $J$ :

(M^{\mathsf T})^{-1}J\,M^{-1}=J

Multiply on the left by $M^{\mathsf T}$ and on the right by $M$ : $J=M^{\mathsf T}JM$ . The two forms imply each other for any invertible $M$ , and running the chain backwards gives the converse. Why physicists keep both forms: $MJM^{\mathsf T}=J$ arises naturally from brackets of new variables (rows of $M$ ); $M^{\mathsf T}JM=J$ arises from Lagrange brackets and from transforming Hamilton's equations. Textbooks pick one and readers transplant formulas — knowing the equivalence proof kills a whole family of sign panics. Verified: matches Goldstein Eqs. (9.55)–(9.66); checked against the lemma chain $J^{-1}=-J$ , $J^2=-I$ , and the inversion formula $M^{-1}=-JM^{\mathsf T}J$ , which reproduces $M^{-1}M=I$ upon substituting the sandwich. Consistency: for $M=I$ (identity map) both forms reduce to $J=J$ .

Which charts carry the fingerprint — and which break it

The theorem is only as useful as one's stock of examples. Three families calibrate the intuition. First, point transformations: change the configuration coordinates alone, $Q=Q(q)$ , and let the momenta come along by the Lagrangian rule $P=(\partial q/\partial Q)\,p$ (in $n$ degrees of freedom, $P_i=\sum_k(\partial q_k/\partial Q_i)p_k$ — the inverse-transpose action). These always pass. This is exactly why the issue is invisible in Lagrangian mechanics: there, any invertible $Q(q)$ is legal, and the momenta silently transform to protect the structure (Derivation C). Second, mutual rescalings: $Q=\lambda q$ , $P=\mu p$ gives $\{Q,P\}=\lambda\mu\,\{q,p\}=\lambda\mu$ . The fingerprint survives iff $\lambda\mu=1$ — you may stretch a coordinate only while compressing its partner, an area-preserving shear of the conjugate plane. The pair $(2q,\,p/2)$ is canonical; $(q,\,2p)$ is not, and the number $\{Q,P\}=2$ is the factor by which conjugate-plane area gets double-counted. Third, the deceptively natural pair everyone reaches for first: position and velocity. With $v=p/m$ ,

\{x,v\}=\Big\{x,\frac{p}{m}\Big\}=\frac{1}{m}\,\{x,p\}=\frac{1}{m}\neq1 \qquad (8)

where: bilinearity pulled the constant $1/m$ straight out. The $(x,v)$ chart — the default of Newtonian intuition — is not canonical (unless $m=1$ in your units). Every bracket formula used blindly in $(x,v)$ silently mis-scales by mass factors; this is the single most common way students break Hamiltonian formulas without noticing.

Go deeper · Derivation C · 2 minPoint transformations automatically preserve the table

Derivation CPoint transformations automatically preserve the table

We are proving that a configuration-space coordinate change $Q_i=Q_i(q)$ , with momenta transformed by the induced rule $P_i=\sum_k A_{ik}(q)\,p_k$ where $A=(\partial q/\partial Q)^{\mathsf T}$ -style inverse Jacobian data, preserves all fundamental brackets — canonicity for free. The strategy is direct evaluation of the three bracket families, using that $Q$ depends only on $q$ and $P$ is linear in $p$ . The one subtlety worth watching is the $\{P_i,P_j\}$ family, where two $q$ -dependent coefficient matrices must cancel against each other through a symmetry of second derivatives.

Step 1. Setup and notation. Let $B_{ik}(q)=\partial Q_i/\partial q_k$ be the configuration Jacobian, and define the momentum rule with its inverse transpose: $P_i=\sum_k (B^{-1})_{ki}\,p_k$ , i.e. $P=(B^{-1})^{\mathsf T}p$ . (Consistency check against one degree of freedom: $B=Q'(q)$ , so $P=p/Q'(q)=(\mathrm dq/\mathrm dQ)\,p$ — the familiar rule.)

Step 2. $\{Q_i,Q_j\}=0$ immediately: both functions depend on $q$ alone, and the bracket pairs a $q$ -derivative of one slot with a $p$ -derivative of the other; every term carries a factor $\partial Q/\partial p=0$ .

Step 3. $\{Q_i,P_j\}$ . Only the term $\sum_k(\partial Q_i/\partial q_k)(\partial P_j/\partial p_k)$ survives ( $Q$ kills the other). With $\partial P_j/\partial p_k=(B^{-1})_{kj}$ :

\{Q_i,P_j\}=\sum_k B_{ik}\,(B^{-1})_{kj}=(BB^{-1})_{ij}=\delta_{ij}

The unit bracket is exactly the statement "the momentum rule inverts the coordinate Jacobian."

Step 4. $\{P_i,P_j\}$ . Both are linear in $p$ with $q$ -dependent coefficients, so the bracket has two terms:

\{P_i,P_j\}=\sum_k\Big(\frac{\partial P_i}{\partial q_k}\frac{\partial P_j}{\partial p_k}-\frac{\partial P_i}{\partial p_k}\frac{\partial P_j}{\partial q_k}\Big)=\sum_{k,l}p_l\Big(\partial_{q_k}(B^{-1})_{li}\,(B^{-1})_{kj}-(B^{-1})_{ki}\,\partial_{q_k}(B^{-1})_{lj}\Big)

Differentiating $B^{-1}$ via $\partial(B^{-1})=-B^{-1}(\partial B)B^{-1}$ and relabeling dummy indices, the two terms become mirror contractions of $\partial^2 Q/\partial q\,\partial q$ — symmetric in its two lower indices — against an antisymmetrized pair. A symmetric object contracted with an antisymmetric pattern vanishes: $\{P_i,P_j\}=0$ . Why the cancellation is not luck: it is the equality of mixed second partials of $Q(q)$ — the same "mixed partials" mechanism that makes Hamiltonian flow symplectic. Whenever a bracket calculation hinges on a miracle, look for a Hessian being symmetric.

Step 5. All three families match table (1); by the fingerprint theorem the point transformation with induced momenta is canonical, for every invertible $Q(q)$ . Contrast: transform $q$ but keep $P=p$ unchanged, and Step 3 gives $\{Q_i,P_j\}=B_{ij}\ne\delta_{ij}$ in general — the momenta must co-transform or the fingerprint smudges. Verified: the 1-DOF specialization $\{Q,P\}=Q'(q)\cdot(1/Q'(q))=1$ checks instantly; the polar-coordinates worked example below confirms the theorem numerically in 2 DOF with all six brackets. Consistent with Goldstein §9.4's point-transformation discussion.

Limiting cases

Case	Result	Meaning
Point transformation $Q=Q(q)$ , $P=(\partial q/\partial Q)p$	table preserved	any configuration relabeling is canonical — Lagrangian freedom survives intact
Mutual rescaling $Q=\lambda q$ , $P=\mu p$	$\{Q,P\}=\lambda\mu$	canonical iff $\lambda\mu=1$ ; otherwise the number reports the area distortion
Position–velocity chart $(x,\,p/m)$	$\{x,v\}=1/m$	the Newtonian default chart is not canonical — mass factors leak into every formula
Coordinate–momentum swap $Q=p$ , $P=-q$	$\{Q,P\}=\{p,-q\}=+1$	canonical: phase space has no intrinsic difference between "position" and "momentum" axes
Time evolution $\eta(0)\to\eta(t)$	$\{q_i(t),p_j(t)\}=\delta_{ij}$	Hamiltonian flow preserves the table at every instant — the flow is a canonical map (proof via Jacobi identity, Algebra of Poisson Brackets)
Quantization $\{\,\}\to[\,]/i\hbar$	$[\hat q_i,\hat p_j]=i\hbar\,\delta_{ij}$	the table survives with $1\to i\hbar$ : the classical fingerprint becomes the canonical commutator

Common trapA smooth invertible relabeling of phase space does not automatically satisfy the table

The trap is treating $\{Q_i,P_j\}=\delta_{ij}$ as a birthright of any reasonable-looking new variables. It is a theorem to check, not a notation to assume. Three specific versions bite hardest. (1) Scaling one variable alone: $(q,2p)$ fails with $\{Q,P\}=2$ , and every "conserved" quantity computed in that chart drifts by the same factor. (2) The physics-default pair $(x,v)$ : off by $1/m$ — brackets computed there are not Poisson brackets. (3) Multi-DOF complacency: in $n\ge2$ degrees of freedom, checking each conjugate pair $\{Q_i,P_i\}=1$ is not enough; the cross-entries $\{Q_i,P_j\}$ ( $i\ne j$ ), $\{Q_i,Q_j\}$ , $\{P_i,P_j\}$ must all vanish too, and coupling between degrees of freedom hides there. The full checking protocol — what to compute, in which order, with which shortcuts — is the subject of Poisson Brackets as a Test for Canonicity.

Worked examples

1 · Rescalings: one passes, one fails (confidence builder)

Check the fundamental brackets for (a) $Q=2q$ , $P=p/2$ and (b) $Q=q$ , $P=2p$ . Interpret both results as areas.

Step 1. Case (a), by bilinearity — no partials needed: $\{Q,P\}=\{2q,\,p/2\}=2\cdot\tfrac12\,\{q,p\}=\mathbf{1}$ . Same-type brackets are trivially zero (one degree of freedom). The table holds: canonical.

Step 2. Case (b): $\{Q,P\}=\{q,2p\}=2\{q,p\}=\mathbf{2}\neq1$ . Fingerprint broken — not canonical.

Step 3. The area reading. The map (a) sends a phase-space rectangle $\mathrm dq\,\mathrm dp$ to $\mathrm dQ\,\mathrm dP=(2\,\mathrm dq)(\tfrac12\mathrm dp)=\mathrm dq\,\mathrm dp$ : stretched horizontally, squeezed vertically, area intact. Map (b) doubles the area: $\mathrm dQ\,\mathrm dP=2\,\mathrm dq\,\mathrm dp$ . The bracket value is the area magnification factor — $\{Q,P\}=\partial(Q,P)/\partial(q,p)$ , the Jacobian determinant, in one degree of freedom.

Step 4. Why you should care operationally: write the harmonic oscillator $H=p^2/2m+\tfrac12 m\omega^2q^2$ in chart (b) by substitution, then naively apply Hamilton's equations in $(Q,P)$ . The predicted frequency comes out wrong by the factor $2$ that the bracket reported. The bracket value is the exact size of the error you commit by pretending the chart is canonical.

2 · Cartesian to polar with induced momenta: the full table, numerically (exam level)

In the plane, define $r=\sqrt{x^2+y^2}$ , $\theta=\arctan(y/x)$ , with induced momenta $p_r=(x p_x+y p_y)/r$ (radial projection) and $p_\theta=x p_y-y p_x$ (angular momentum). At the phase point $x=3$ , $y=4$ , $p_x=2$ , $p_y=1$ (SI units), verify numerically all six independent fundamental brackets.

Step 1. Values at the point: $r=\sqrt{9+16}=5$ , $\theta=\arctan(4/3)=0.9273$ , $p_r=(3\cdot2+4\cdot1)/5=10/5=2.000$ , $p_\theta=3\cdot1-4\cdot2=-5.000$ .

Step 2. Partial derivatives, evaluated numerically. For $r$ : $\partial r/\partial x=x/r=0.600$ , $\partial r/\partial y=y/r=0.800$ , momentum partials zero. For $\theta$ : $\partial\theta/\partial x=-y/r^2=-0.160$ , $\partial\theta/\partial y=x/r^2=0.120$ , momentum partials zero. For $p_r$ (write $S=xp_x+yp_y=10$ ): $\partial p_r/\partial x=p_x/r-Sx/r^3=0.400-0.240=0.160$ , $\partial p_r/\partial y=p_y/r-Sy/r^3=0.200-0.320=-0.120$ , $\partial p_r/\partial p_x=x/r=0.600$ , $\partial p_r/\partial p_y=y/r=0.800$ . For $p_\theta$ : $\partial p_\theta/\partial x=p_y=1$ , $\partial p_\theta/\partial y=-p_x=-2$ , $\partial p_\theta/\partial p_x=-y=-4$ , $\partial p_\theta/\partial p_y=x=3$ .

Step 3. The two conjugate brackets:

\{r,p_r\}=(0.600)(0.600)+(0.800)(0.800)=0.360+0.640=\mathbf{1.000}

\{\theta,p_\theta\}=(-0.160)(-4)+(0.120)(3)=0.640+0.360=\mathbf{1.000}

Step 4. The four cross brackets. $\{r,\theta\}=0$ outright (neither depends on momenta). $\{r,p_\theta\}=(0.600)(-4)+(0.800)(3)=-2.400+2.400=\mathbf{0}$ . $\{\theta,p_r\}=(-0.160)(0.600)+(0.120)(0.800)=-0.096+0.096=\mathbf{0}$ . Finally the momentum–momentum bracket, which has all four term pairs:

\{p_r,p_\theta\}=(0.160)(-4)-(0.600)(1)+(-0.120)(3)-(0.800)(-2)=-0.640-0.600-0.360+1.600=\mathbf{0}

Step 5. All six brackets match table (1) exactly — polar coordinates with their induced momenta are a canonical chart, as Derivation C guarantees for every point transformation. Notice what made the last bracket vanish: four terms of order unity conspiring to zero, the numerical shadow of the mixed-partials cancellation in Derivation C, Step 4. And notice what would have failed: had we naively kept $p_\theta$ but used the "velocity-like" $\dot\theta$ instead, every $\theta$ -row bracket would inherit a factor $1/(mr^2)$ — the $(x,v)$ disease of equation (8) in angular clothing.

Origins · 1 minPaul Dirac and how this was discovered

Paul Dirac

1902–1984 · Cambridge

Dirac was a research student at Cambridge in the autumn of 1925 when Heisenberg's first quantum paper reached him, with its strange arrays that multiplied in order-dependent fashion. Where others saw a defect, Dirac hunted for the classical ancestor of the non-commutativity — and recognized that the commutator $\hat f\hat g-\hat g\hat f$ behaves grammatically like a Poisson bracket: bilinear, antisymmetric, Leibniz in each slot. The consequence, worked out in his 1925 paper "The Fundamental Equations of Quantum Mechanics," is that quantum mechanics need not invent its algebra: it inherits the classical one, with the fundamental table $\{q,p\}=1$ transcribed as $[\hat q,\hat p]=i\hbar$ . That one substitution is why the table on this page appears on page one of quantum textbooks. Dirac's taste — find the load-bearing structure, discard the scaffolding — runs through everything he touched: the transformation theory uniting Heisenberg's and Schrödinger's mechanics, the relativistic electron equation, antimatter, magnetic monopoles, and the constrained-systems generalization now called the Dirac bracket, which extends this page's table to systems where the naive one fails. He held Newton's old chair, the Lucasian professorship, for 37 years and wrote the field's most austere and durable textbook.

History

1809

— Poisson introduces the bracket in a memoir on planetary perturbations; the coordinate brackets appear as bookkeeping identities

1842

— Jacobi's Königsberg lectures organize transformation theory: preserving the brackets emerges as the mark of a canonical change of variables

1890

— Poincaré's integral invariants tie the preserved brackets to conserved phase-space areas

1925

— Dirac maps the table to commutators,

\{q,p\}=1\to[\hat q,\hat p]=i\hbar

: canonical quantization is born

1930

— Dirac's Principles of Quantum Mechanics codifies the bracket-to-commutator dictionary as the field's standard grammar

1950

— Dirac's constrained-Hamiltonian theory generalizes the table to gauge systems via the Dirac bracket

Going further

References. Goldstein, Poole & Safko, Classical Mechanics, 3rd ed., §9.4–9.5 · Landau & Lifshitz, Mechanics, 3rd ed., §42 · Morin, Introduction to Classical Mechanics, Ch. on Hamiltonian mechanics · Taylor, Classical Mechanics, Ch. 13 · Arnold, Mathematical Methods of Classical Mechanics, §40

Provenance. Bracket definition and fundamental brackets from Goldstein §9.4–9.5 and Landau–Lifshitz §42. Symplectic equivalence ( $MJM^{\mathsf T}=J$ ) derived and cross-checked against Goldstein Eqs. (9.55)–(9.66). Time-evolution invariance uses the Jacobi identity (Goldstein §9.6). Verifier should recompute: (1) the three fundamental brackets from the definition; (2) the chain-rule identity $\{\zeta_a,\zeta_b\}=(MJM^{\mathsf T})_{ab}$ ; (3) that $J^{\mathsf T}=-J$ , $J^2=-I$ , $J^{-1}=-J$ ; (4) $\{p,-q\}=+1$ and $\{2q,2p\}=4$ . All numbers ( $\delta_{ij}$ , factor 4, unit bracket) are exact and standard. The polar-chart numerical table ( $x=3$ , $y=4$ , $p_x=2$ , $p_y=1$ : all six brackets equal to $1,1,0,0,0,0$ ) was computed for this article and cross-checked by explicit partial derivatives.

Builds on: Poisson Brackets: Definition · Definition of Canonical Transformations · Algebra of Poisson Brackets Leads to: Poisson Brackets as a Test for Canonicity · Symplectic Condition · Poisson-Bracket Formulation of Dynamics · Symplectic Group and Phase-Space Geometry

Are the fundamental brackets a definition or a theorem?

Both, depending on where you stand. Starting from the coordinate definition of the bracket (as in Poisson Brackets: Definition), the table is a two-line theorem — evaluate the definition on coordinate functions and Kronecker deltas collapse the sums. But in the modern structural view the logic runs backwards: one defines a canonical chart as any coordinate system satisfying the table, because Derivation A shows the table alone (plus grammar) reconstructs every bracket. The second stance is the one quantum mechanics inherits: there is no "definition" of the commutator to compute from — the table $[\hat q,\hat p]=i\hbar$ is imposed as the founding axiom, precisely because classically it carries the whole structure.

Why is the bracket of q1 with p2 zero — surely the dynamics can couple different degrees of freedom?

You are mixing two different couplings. The fundamental brackets encode the kinematic wiring of phase space — which variable is conjugate to which — and that wiring is fixed before any Hamiltonian is chosen. Dynamical coupling lives in $H$ : a term like $q_1q_2$ in the Hamiltonian makes $\dot p_1$ depend on $q_2$ , but it does so through the bracket machinery, not by rewiring it. The clean separation is visible in the flow reading: $\{q_1,p_2\}=0$ says the translation generated by $p_2$ moves $q_2$ only, leaving $q_1$ untouched — true regardless of how violently the two oscillators later exchange energy under the dynamics.

If I make my variables dimensionless the unit bracket picks up dimensions — have I broken canonicity?

The bracket $\{q,p\}$ carries dimensions of $[q][p]/([q][p])=1$ … more carefully: $\{f,g\}$ has the dimensions of $fg$ divided by action, because one $\partial/\partial q$ pairs with one $\partial/\partial p$ . So $\{q,p\}=1$ is "1 per unit action-per-action" — dimensionless — and rescaling to dimensionless variables $\tilde q=q/q_0$ , $\tilde p=p/p_0$ gives $\{\tilde q,\tilde p\}=1/(q_0p_0)$ . That is an honest fingerprint violation unless you also rescale time/energy so that $q_0p_0$ is your new unit of action — the standard trick is choosing $q_0p_0=\hbar$ or $=1$ in natural units, making the rescaling canonical-with-rescaled-Hamiltonian. The practical rule: nondimensionalize in conjugate pairs whose scale product is your action unit, and the table survives intact.

Is the position–velocity chart really unusable? Physics cannot depend on whether I call it p or mv.

Physics does not depend on it — formulas do. In $(x,v)$ the honest bracket structure is $\{x,v\}=1/m$ : a perfectly legitimate noncanonical bracket, and you can do correct mechanics with it if you carry the $1/m$ through every formula (the general machinery is a Poisson tensor $\Pi$ replacing $J$ ). What you may not do is transplant canonical-chart formulas — $\dot f=\{f,H\}$ with the standard definition, generating functions, quantization rules — into $(x,v)$ unmodified; each acquires mass factors. The canonical formalism's bargain is: use conjugate momenta, and every formula is chart-universal. Velocity buys intuition and costs universality.

Do the fundamental brackets change in curvilinear coordinates, like spherical?

No — provided the momenta are the conjugate ones, and that is the entire content of Derivation C. Spherical coordinates $(r,\theta,\phi)$ with the Lagrangian-induced momenta $(p_r,p_\theta,p_\phi)$ satisfy the table exactly, as the polar worked example verifies numerically in 2D. What changes in curvilinear charts is the Hamiltonian's appearance (e.g. $p_\theta^2/2mr^2$ terms), never the bracket table. The table smudges only when you pair coordinates with non-conjugate partners: $\theta$ with $L_z$ is fine ( $\{\theta_{\text{azimuthal}},L_z\}=1$ — indeed $L_z=p_\phi$ is the conjugate momentum), but $\theta$ with $m\dot\theta$ is not.

Why does the classical 1 become i hbar rather than just hbar — where does the i come from?

Hermiticity. Observables quantize to Hermitian operators, and the commutator of two Hermitian operators is anti-Hermitian: $([\hat A,\hat B])^\dagger=-[\hat A,\hat B]$ . To equate it with something classical-real you must strip the anti-Hermiticity, and dividing by $i$ does exactly that — $[\hat q,\hat p]/i\hbar$ is Hermitian-compatible with the real number $\{q,p\}=1$ . The $\hbar$ is then forced by dimensions: a commutator of position with momentum carries action units, the classical bracket is dimensionless (see the dimensionless-variables FAQ), so the conversion constant must carry action — and experiment fixes it to be Planck's. Deeper treatments (Moyal/deformation quantization) derive the same $i\hbar$ as the first-order coefficient of a systematic expansion.

Can two variables satisfy the conjugate-pair bracket of 1 while other brackets in the set fail?

Yes, and this is the standard multi-DOF trap. Take $Q_1=q_1$ , $P_1=p_1+q_2$ , $Q_2=q_2$ , $P_2=p_2$ : each diagonal pair gives $\{Q_i,P_i\}=1$ , but $\{P_1,Q_2\}=\{p_1+q_2,q_2\}=0$ ✓ while $\{P_1,P_2\}=\{p_1+q_2,p_2\}=\{q_2,p_2\}=1\ne0$ ✗ — the momenta no longer commute with each other, and the chart is not canonical despite two perfect-looking conjugate pairs. The fingerprint is the whole matrix $\{\eta_a,\eta_b\}=J_{ab}$ , all $2n\times2n$ entries. The efficient checking order (and which entries symmetry lets you skip) is part of the algorithm in Poisson Brackets as a Test for Canonicity.

Is there anything like the fundamental brackets for systems with constraints, where q and p are not independent?

Yes — this is exactly where Dirac's later work enters. When constraints $\phi_a(q,p)=0$ carve out a submanifold (gauge theories, systems with redundant variables), the naive table can be inconsistent with the constraints, and Dirac's construction replaces the Poisson bracket by the Dirac bracket, built by projecting out the constraint directions. On the constraint surface the Dirac brackets of a well-chosen reduced set of variables again take the fundamental form — a new, smaller table — and quantization proceeds by transcribing that table into commutators. The moral survives all generalizations: identify the honest conjugate pairs first; the table is the structure everything else hangs on.