Poisson Brackets: Definition

In September 1925, a young Paul Dirac was out on his usual Sunday walk when a half-remembered structure from classical mechanics surfaced: those Poisson brackets — didn't they combine exactly like the strange non-commuting products in Heisenberg's new paper? The libraries were closed; he had to wait until Monday morning to check. The definition he looked up is the subject of this page: for any two functions $f$ and $g$ on phase space, $\{f,g\}=\sum_i(\partial_{q_i}f\,\partial_{p_i}g-\partial_{p_i}f\,\partial_{q_i}g)$ — a century-old bookkeeping device from planetary perturbation theory that turned out to be the skeleton quantum mechanics would hang on. Its classical job is already remarkable. The bracket answers one question with total generality: how fast does the quantity $f$ change when the system flows along the dynamics that $g$ generates? Set $g=H$ and you get time evolution of anything — $\dot f=\{f,H\}+\partial f/\partial t$ — with Hamilton's equations as the special case $f=q_i$ or $p_i$ . The entire content of the symplectic geometry you met in Symplectic Condition compresses into a three-line multiplication table, the fundamental brackets $\{q_i,p_j\}=\delta_{ij}$ , and the bracket's value is the same in every canonical chart — which is why it can serve as the definitive test of canonicity. This page owns the definition and its immediate consequences: symbol-by-symbol meaning, the matrix form, evolution, invariance, and two fully numeric computations. The algebraic laws the bracket obeys — Leibniz, Jacobi, the Lie-algebra structure — get their own treatment in Algebra of Poisson Brackets.

Key takeaways

The Poisson bracket of two phase-space functions is $\{f,g\}=\sum_i\left(\frac{\partial f}{\partial q_i}\frac{\partial g}{\partial p_i}-\frac{\partial f}{\partial p_i}\frac{\partial g}{\partial q_i}\right)$ — bilinear, antisymmetric, built from first derivatives only.
Physical meaning: $\{f,g\}$ is the directional derivative of $f$ along the Hamiltonian flow generated by $g$ . When $g=H$ , this flow is the physical time evolution.
The fundamental brackets $\{q_i,p_j\}=\delta_{ij}$ , $\{q_i,q_j\}=\{p_i,p_j\}=0$ are the entire symplectic structure written as a multiplication table.
Any observable evolves by $\frac{\mathrm{d}f}{\mathrm{d}t}=\{f,H\}+\frac{\partial f}{\partial t}$ ; Hamilton's equations are the cases $f=q_i,\,p_i$ .
The bracket's value is invariant under canonical transformations — computed in any canonical chart, it is the same number; this is the bracket test of canonicity in embryo.
Dirac's quantization dictionary $\{f,g\}\to\frac{1}{i\hbar}[\hat f,\hat g]$ makes the bracket the classical skeleton of quantum commutators.
Key idea: the Poisson bracket turns phase space into a Lie algebra — $\{f,g\}$ is the rate of change of $f$ under the flow generated by $g$ , and every piece of Hamiltonian mechanics is a statement about brackets.

The core insight

The Poisson bracket turns phase space into a Lie algebra: $\{f,g\}$ is the rate of change of $f$ under the flow generated by $g$ , and every piece of Hamiltonian mechanics — evolution, conservation, canonicity — is a statement about brackets.

Where this lives today

The Poisson-bracket formalism is foundational in modern computational physics — most visibly where conservation laws must survive discretization. Plasma physics rebuilt its master equations in bracket form: ideal magnetohydrodynamics and modern gyrokinetic theory formulate plasma dynamics as noncanonical Hamiltonian systems, written as $\dot F=\{F,H\}$ with generalized (Morrison–Greene) brackets whose degenerate directions encode magnetic-flux and entropy conservation; simulation codes for ITER-class devices preserve those invariants by discretizing the bracket itself rather than the equations. Geophysical fluid dynamics runs the same play — potential-vorticity conservation, the backbone quantity of weather and climate models, is a bracket Casimir. And the quantization dictionary is now engineering practice in reverse: quantum-computing hardware teams designing superconducting circuits start from the circuit's classical Lagrangian, extract canonical pairs of flux and charge, write the fundamental brackets, and then promote $\{\Phi,Q\}=1$ to $[\hat\Phi,\hat Q]=i\hbar$ — Dirac's 1925 Sunday-walk correspondence, executed daily as step one of qubit design.

The physical picture

Every smooth function $g$ on phase space is secretly an engine: it generates a Hamiltonian vector field $X_g=\left(\frac{\partial g}{\partial p},\ -\frac{\partial g}{\partial q}\right)$ , whose integral curves define a flow. Since $X_g\cdot\nabla g=0$ , the motion is tangent to the level curves of $g$ , never across them. The momentum function generates the flow "translate in position"; angular momentum generates "rotate"; the Hamiltonian generates the system's time evolution. This is the infinitesimal-canonical-transformation machinery of Type-2 Generating Function F2(q,P,t), now viewed from the observables' side. The bracket is the meter attached to the engine. $\{f,g\}$ reads off how fast the quantity $f$ ticks when the system is pushed by $g$ 's flow:

Figure

The bracket as a flow meter — g generates motion along its own level curves, and the bracket reads how fast f changes along that motion

Antisymmetry, $\{f,g\}=-\{g,f\}$ , is visible in the picture: the Hamiltonian flow generated by $g$ is tangent to its own level sets, so $g$ remains constant along its own trajectories — $\{g,g\}=0$ — and "how $f$ changes under $g$ " must be the exact negative of "how $g$ changes under $f$ ." Set $g=H$ and $\{H,H\}=0$ is conservation of energy for autonomous systems: the dynamics cannot climb its own generator. The second picture to internalize is the multiplication table. Every bracket of every pair of observables, however complicated, reduces by algebra to the three atomic statements $\{q_i,p_j\}=\delta_{ij}$ , $\{q_i,q_j\}=0$ , $\{p_i,p_j\}=0$ — the symplectic geometry of phase space compressed into one line. Quantum mechanics keeps the table and changes only the currency: $\{q,p\}=1$ becomes $[\hat q,\hat p]=i\hbar$ . When physicists say classical and quantum mechanics "share a skeleton," this table is the skeleton.

Figure

The atomic multiplication table — every bracket reduces to three entries, and quantum mechanics keeps the table while changing the currency

The definition, read symbol by symbol

\{f,g\}=\sum_{i=1}^{n}\left(\frac{\partial f}{\partial q_i}\frac{\partial g}{\partial p_i}-\frac{\partial f}{\partial p_i}\frac{\partial g}{\partial q_i}\right) \qquad (1)

where: $f,g$ — smooth phase-space functions, possibly explicitly time-dependent; $q_i,p_i$ — canonical pairs, $i=1,\dots,n$ ; each partial derivative holds all other phase-space variables fixed. Read it physically: the first product asks how $f$ responds to position changes times how strongly $g$ pushes positions (through $\partial g/\partial p$ , since momenta drive coordinates); the second, with the minus, does the mirrored job for momenta. Note the first-derivatives-only structure — the bracket of two observables is again an observable, ready for further bracketing, which is what makes an algebra possible. Evaluating (1) on the coordinates themselves collapses the sum by Kronecker deltas (Derivation A):

\{q_i,p_j\}=\delta_{ij},\qquad \{q_i,q_j\}=0,\qquad \{p_i,p_j\}=0 \qquad (2)

where: $\delta_{ij}$ — 1 for $i=j$ , else 0. These fundamental brackets are exactly the entries of the symplectic matrix, which becomes explicit in the compact form

\{f,g\}=\left(\nabla_{z}f\right)^{\mathsf T}J\,\nabla_{z}g,\qquad z=(q,p)^{\mathsf T},\quad J=\begin{pmatrix}0&I_n\\-I_n&0\end{pmatrix} \qquad (3)

where: $\nabla_z$ — the $2n$ -gradient; $J$ — the same antisymmetric form that writes Hamilton's equations as $\dot z=J\nabla_zH$ . One geometry, three notations: matrix ( $J$ ), calculus (the bracket), and, in the language of Symplectic Group and Phase-Space Geometry, the two-form $\omega$ . The basic algebraic properties of the Poisson bracket follow directly from definition (1); the complete algebraic structure (including the Jacobi identity) is developed in Algebra of Poisson Brackets:

\{f,g\}=-\{g,f\},\qquad \{f,gh\}=\{f,g\}h+g\{f,h\},\qquad \{f,c\}=0\ \ (c\ \text{const}) \qquad (4)

where: antisymmetry makes self-brackets vanish; the Leibniz rule says "bracket with $f$ " acts like a derivative — because it is one, the derivative along $f$ 's flow.

Derivation AFundamental brackets and antisymmetry from the bare definition

We are establishing the multiplication table (2) and the sign structure (4) directly from definition (1) — the two facts every other bracket computation reduces to. The strategy is plain evaluation: coordinates and momenta are the simplest possible phase-space functions, with partial derivatives that are pure Kronecker deltas, so the double sum collapses. Watch which term of the definition survives in each case; that pattern is the bracket's anatomy in miniature.

Step 1. Partial derivatives of the coordinate functions themselves: $\partial q_i/\partial q_k=\delta_{ik}$ and $\partial q_i/\partial p_k=0$ (the phase-space variables are independent); likewise $\partial p_j/\partial p_k=\delta_{jk}$ , $\partial p_j/\partial q_k=0$ .

Step 2. Insert into the definition for $\{q_i,p_j\}$ :

\{q_i,p_j\}=\sum_{k}\left(\delta_{ik}\,\delta_{jk}-0\cdot0\right)=\delta_{ij}

Only the first product survives, and the deltas lock $k$ to both $i$ and $j$ simultaneously.

Step 3. For $\{q_i,q_j\}$ every term carries a factor $\partial q/\partial p=0$ ; for $\{p_i,p_j\}$ every term carries $\partial p/\partial q=0$ . Both vanish identically — same-type variables never see each other through the bracket.

Step 4. Antisymmetry in general: swap $f\leftrightarrow g$ in (1),

\{g,f\}=\sum_i\left(\frac{\partial g}{\partial q_i}\frac{\partial f}{\partial p_i}-\frac{\partial g}{\partial p_i}\frac{\partial f}{\partial q_i}\right)=-\{f,g\}

since reordering ordinary products and factoring $-1$ restores the original expression negated. Why the table is "the whole geometry": equation (3) shows the bracket is built from $J$ alone, and (2) says the coordinate functions read out $J$ 's entries exactly. Knowing the bracket on coordinates plus bilinearity and Leibniz determines it on all polynomials, hence (by density) on everything. Consistency check: the matrix form (3) evaluated on $z_i,z_j$ returns $J_{ij}$ , which is precisely the fundamental table.

Evolution: the bracket with H is the time derivative

\frac{\mathrm{d}f}{\mathrm{d}t}=\{f,H\}+\frac{\partial f}{\partial t} \qquad (5)

where: $\mathrm{d}f/\mathrm{d}t$ — total rate of change along the actual trajectory; $\{f,H\}$ — the dynamical part, transport by the flow; $\partial f/\partial t$ — only the explicit clock dependence of $f$ 's formula (zero for observables like $p^2$ or $xp_y$ ). This single equation is Hamiltonian mechanics: it contains Hamilton's equations (set $f=q_i$ or $p_i$ and compare with the fundamental brackets), it defines conservation ( $f$ conserved $\iff\{f,H\}=0$ for time-independent $f$ — the doorway to Constants of Motion via Poisson Brackets), and after Dirac's substitution it becomes the Heisenberg equation of motion.

Go deeper · Derivation B · 1 minThe evolution equation from the chain rule

Derivation BThe evolution equation from the chain rule

We derive the fundamental evolution equation (5): the total time derivative of any observable equals its Poisson bracket with the Hamiltonian plus its explicit time dependence. The strategy is the two-line classic — chain rule along the trajectory, then Hamilton's equations substituted — but the payoff sentence matters: once (5) exists, you can do dynamics without ever solving for trajectories, manipulating observables algebraically instead. That shift of viewpoint is what survives into quantum mechanics.

Step 1. Chain rule for $f\big(q(t),p(t),t\big)$ along any curve in phase space:

\frac{\mathrm{d}f}{\mathrm{d}t}=\sum_i\left(\frac{\partial f}{\partial q_i}\,\dot q_i+\frac{\partial f}{\partial p_i}\,\dot p_i\right)+\frac{\partial f}{\partial t}

Step 2. Make the curve a trajectory: insert Hamilton's canonical equations $\dot q_i=\partial H/\partial p_i$ , $\dot p_i=-\partial H/\partial q_i$ (see Hamilton's Canonical Equations — used here, not re-derived):

\frac{\mathrm{d}f}{\mathrm{d}t}=\sum_i\left(\frac{\partial f}{\partial q_i}\frac{\partial H}{\partial p_i}-\frac{\partial f}{\partial p_i}\frac{\partial H}{\partial q_i}\right)+\frac{\partial f}{\partial t}

Step 3. Recognize definition (1) with $g=H$ :

\frac{\mathrm{d}f}{\mathrm{d}t}=\{f,H\}+\frac{\partial f}{\partial t}

Why the two time derivatives differ: $\partial f/\partial t$ asks how $f$ 's formula changes at a frozen phase point; $\mathrm{d}f/\mathrm{d}t$ asks how $f$ 's value changes for the moving system. The Poisson bracket is precisely the contribution arising from motion through phase space, while $\partial f/\partial t$ accounts for any explicit time dependence.

Step 4. Close the loop: set $f=q_i$ , use the fundamental brackets to get $\dot q_i=\{q_i,H\}=\partial H/\partial p_i$ , and similarly $\dot p_i=\{p_i,H\}=-\partial H/\partial q_i$ . Hamilton's equations are the evolution equation's own special case — the formalism has swallowed its starting point. Consistency check: setting $f=H$ time-independent gives $\dot H=\{H,H\}=0$ — energy conservation, as antisymmetry demands.

Invariance: the bracket does not care about your chart

Here is the property that elevates the bracket from convenient notation to structural object: its value is chart-independent across canonical coordinates.

\{f,g\}_{q,p}=\{f,g\}_{Q,P}\qquad\text{for every canonical }(q,p)\to(Q,P) \qquad (6)

where: the subscripts name which variables the partial derivatives run over. Physical statements built from brackets — "this quantity is conserved," "these two observables generate commuting flows" — therefore hold in every canonical chart at once, which is what a physical statement should do.

Go deeper · Derivation C · 1 minBracket invariance under canonical transformations

Derivation CBracket invariance under canonical transformations

We are proving equation (6): compute the bracket of the same two functions in any canonical chart and the number agrees. The strategy uses the matrix form (3) plus the gradient transformation law, and the punchline is that invariance holds precisely because the Jacobian satisfies the symplectic condition — making "preserves brackets" and "is canonical" the same property viewed from two sides. This is the logical foundation of the canonicity test developed in the companion article.

Step 1. Write the bracket in the old chart in matrix form: $\{f,g\}_{z}=(\nabla_zf)^{\mathsf T}J\,\nabla_zg$ with $z=(q,p)$ .

Step 2. Transform the gradients. With $Z=(Q,P)$ , $M_{ab}=\partial Z_a/\partial z_b$ , the chain rule gives $\nabla_zf=M^{\mathsf T}\nabla_Zf$ — gradients pull back through the transpose (covariant), as in Symplectic Condition.

Step 3. Substitute both gradients:

\{f,g\}_{z}=(\nabla_Zf)^{\mathsf T}\,M J M^{\mathsf T}\,\nabla_Zg

Step 4. Invoke canonicity: $MJM^{\mathsf T}=J$ . Then

\{f,g\}_{z}=(\nabla_Zf)^{\mathsf T}J\,\nabla_Zg=\{f,g\}_{Z}

Why this is an equivalence, not a one-way street: Step 3 holds for any invertible map; the final equality holds for all $f,g$ iff the sandwich equals $J$ . So bracket-invariance fails exactly where the symplectic condition fails — compute $\{Q,P\}$ in the old chart and any deviation from 1 is a direct reading of how non-canonical the map is (recall the factor-2 impostor of the symplectic-condition article). Consistency check: evaluating (6) on the coordinate functions themselves reproduces the statement that canonical maps preserve the fundamental brackets.

Limiting cases

Case	Result	Meaning
Coordinates and momenta	$\{q_i,p_j\}=\delta_{ij}$ , others zero	the multiplication table; all of symplectic geometry in one line
Second slot $=H$	$\{f,H\}=\dot f-\partial f/\partial t$	brackets with the Hamiltonian are time evolution
Second slot $=p_j$	$\{f,p_j\}=\partial f/\partial q_j$	momentum generates translation; bracket = spatial derivative
Second slot $=L_z$	$\{f,L_z\}$ = rate of change of $f$ under rotation about $z$	angular momentum generates rotations
Equal arguments	$\{f,f\}=0$	a flow preserves its own generator — energy conservation as algebra
Constant argument	$\{f,c\}=0$	constants are transparent to every flow

Common trapTwo sign conventions circulate — mixing them silently negates your physics

The convention here (Goldstein, Landau–Lifshitz) puts $\partial f/\partial q$ first: $\{f,g\}=\sum(\partial_qf\,\partial_pg-\partial_pf\,\partial_qg)$ , giving $\{q,p\}=+1$ and $\dot f=\{f,H\}$ . A minority of texts define the bracket with the opposite overall sign, which flips every bracket: $\{q,p\}=-1$ and evolution becomes $\dot f=\{H,f\}$ . Both are internally consistent; a formula transplanted between conventions is wrong by a global minus — conserved quantities stay conserved, but every rate, torque, and precession direction reverses. Before using a formula from another source, first check its sign convention by evaluating the fundamental bracket $\{q,p\}$ . Same warning one level up: the quantum dictionary is $\{f,g\}\to[\hat f,\hat g]/i\hbar$ with this convention; flip the classical sign and the $i$ moves too.

Worked examples

1 · A vertically thrown mass, run entirely through brackets (confidence builder)

A mass $m=2.0\ \mathrm{kg}$ moves vertically in gravity: $H=\dfrac{p^{2}}{2m}+mgz$ with $g=9.8\ \mathrm{m/s^2}$ . At the phase point $z=5.0\ \mathrm{m}$ , $p=+6.0\ \mathrm{kg\,m/s}$ (rising), compute (a) $\dot z$ and $\dot p$ from brackets, (b) the rate of change of the observable $f=zp$ , checked two ways.

Step 1. Compute the coordinate's time evolution using the Poisson bracket: $\{z,H\}=\dfrac{\partial z}{\partial z}\dfrac{\partial H}{\partial p}-\dfrac{\partial z}{\partial p}\dfrac{\partial H}{\partial z}=\dfrac{p}{m}-0=3.0\ \mathrm{m/s}$ . So $\dot z=3.0\ \mathrm{m/s}$ — the velocity, read off without touching Newton.

Step 2. The momentum's bracket: $\{p,H\}=0-1\times mg=-mg=-19.6\ \mathrm{N}$ . So $\dot p=-19.6\ \mathrm{N}$ — the weight, with its sign.

Step 3. Now the composite observable $f=zp$ (no explicit time dependence). Direct from the definition:

\{zp,H\}=\frac{\partial(zp)}{\partial z}\frac{\partial H}{\partial p}-\frac{\partial(zp)}{\partial p}\frac{\partial H}{\partial z}=p\cdot\frac{p}{m}-z\cdot mg=\frac{36}{2.0}-5.0\times19.6=18.0-98.0=\mathbf{-80.0\ J}

Step 4. Cross-check by leaning on Steps 1–2 and the product rule: $\dfrac{\mathrm{d}}{\mathrm{d}t}(zp)=\dot z\,p+z\,\dot p=3.0\times6.0+5.0\times(-19.6)=18.0-98.0=-80.0\ \mathrm{J}$ ✓. (Units: $zp$ carries $\mathrm{J\,s}$ , so its rate is joules.) The quantity $zp$ — the one-dimensional virial — is currently decreasing: the mass is high and decelerating, and the gravitational term dominates the kinetic one.

2 · When is angular momentum conserved? Ask the bracket (exam level)

A particle of mass $m$ moves in the anisotropic two-dimensional potential $V=\tfrac{k}{2}\left(x^{2}+4y^{2}\right)$ with $k=2.0\ \mathrm{N/m}$ . (a) Compute $\{L_z,H\}$ symbolically. (b) Evaluate the rate $\dot L_z$ at $x=0.50\ \mathrm{m}$ , $y=0.40\ \mathrm{m}$ and cross-check against the torque. (c) Confirm conservation in the isotropic limit.

Step 1. Setup. $H=\dfrac{p_x^{2}+p_y^{2}}{2m}+V(x,y)$ and $L_z=xp_y-yp_x$ . Split the bracket: $\{L_z,H\}=\{L_z,T\}+\{L_z,V\}$ .

Step 2. Kinetic part. $\{L_z,T\}$ needs $\partial L_z/\partial x=p_y$ , $\partial L_z/\partial y=-p_x$ , $\partial T/\partial p_x=p_x/m$ , $\partial T/\partial p_y=p_y/m$ , and $\partial L_z/\partial p_x=-y$ , $\partial L_z/\partial p_y=x$ against $\partial T/\partial x=\partial T/\partial y=0$ :

\{L_z,T\}=p_y\frac{p_x}{m}+(-p_x)\frac{p_y}{m}-0=0

Kinetic energy is rotationally symmetric — it never torques.

Step 3. Potential part. $V$ has no momentum dependence, so only the second half of the definition survives:

\{L_z,V\}=-\frac{\partial L_z}{\partial p_x}\frac{\partial V}{\partial x}-\frac{\partial L_z}{\partial p_y}\frac{\partial V}{\partial y}=+y\,(kx)-x\,(4ky)=kxy-4kxy=\mathbf{-3kxy}

Step 4. Numbers: $\dot L_z=\{L_z,H\}=-3kxy=-3\times2.0\times0.50\times0.40=\mathbf{-1.20\ N\,m}$ . Cross-check with elementary mechanics: the force is $\mathbf F=-\nabla V=(-kx,\,-4ky)$ , and the torque about $z$ is $\tau_z=xF_y-yF_x=x(-4ky)-y(-kx)=-4kxy+kxy=-3kxy=-1.20\ \mathrm{N\,m}$ ✓. The Poisson bracket reproduces the torque directly: $\dot L_z=\{L_z,H\}=\tau_z$ , without ever introducing the vector cross product explicitly.

Step 5. Isotropic limit: replace $4y^2$ by $y^2$ . Then $\{L_z,V\}=kxy-kxy=0$ : angular momentum is conserved exactly when the potential is rotationally symmetric — the bracket has re-derived a Noether statement by pure differentiation. The anisotropy factor 4 is the entire obstruction, and its "strength" $-3kxy$ vanishes on the axes (where the force is radial even in the anisotropic trap) — a detail the symmetry argument alone would not hand you.

Origins · 1 minSiméon Denis Poisson and how this was discovered

Siméon Denis Poisson

1781–1840 · École Polytechnique, Paris

Poisson introduced his bracket in an 1809 memoir on planetary perturbations, as a computational device for a concrete headache: given two known constants of a perturbed orbital motion, how do the slowly drifting orbital elements interact? Buried in the memoir was the theorem now bearing his name — the bracket of two constants of motion is again a constant — which Jacobi would later single out as Poisson's deepest result and rebuild into the identity that anchors the bracket algebra. The bracket was characteristic of the man: a Polytechnique prodigy (student of Lagrange and Laplace, professor there by 25) who declared that "life is good for only two things — doing mathematics and teaching it," and whose name now attaches to an equation (electrostatics), a distribution (statistics), a ratio (elasticity), and a spot (optics — the bright point he predicted as an objection to Fresnel's wave theory, only for Arago to promptly observe it). He produced over 300 works of relentless, methodical calculation. The bracket outgrew him twice: once when Jacobi and Lie made it geometry, and again in September 1925, when Dirac spent an anxious Sunday waiting for a library to open so he could confirm that Poisson's planetary bookkeeping was the exact classical shadow of Heisenberg's quantum products.

History

1809

— Poisson's memoir on planetary perturbations introduces the bracket and the constants-of-motion theorem

1842

— Jacobi, in his Königsberg lectures, elevates the bracket: the Jacobi identity and the algebra take shape

1870

— Lie builds continuous transformation groups on infinitesimal generators — brackets as the algebra of flows

1890

— Poincaré's integral invariants tie the bracket's invariance to the geometry of phase space

1925

— Dirac identifies the commutator-to-bracket correspondence, founding canonical quantization

1946

— Groenewold and Moyal sharpen the dictionary: the Moyal bracket deforms Poisson's in powers of Planck's constant

Going further

References. Goldstein, Poole & Safko, Classical Mechanics (3rd ed.), §9.5 · Landau & Lifshitz, Mechanics (Vol. 1, 3rd ed.), §42 · Morin, Introduction to Classical Mechanics, Ch. 15 (Hamiltonian formalism) · Taylor, Classical Mechanics, Ch. 13 · Arnold, Mathematical Methods of Classical Mechanics, §40 Key results to keep in mind: the fundamental brackets $\{q_i,p_j\}=\delta_{ij}$ follow directly from the definition; the evolution equation $\dot f=\{f,H\}+\partial_t f$ follows from the chain rule with Hamilton's equations; antisymmetry, Leibniz, and Jacobi all follow from the definition; and canonical invariance rests on the symplectic condition $MJM^\top=J$ . The numeric examples ( $zp$ evolution at $z=5.0$ , $p=6.0$ with $m=2.0$ ; anisotropic-trap torque $-3kxy$ at $x=0.50$ , $y=0.40$ , $k=2.0$ ) were each cross-checked by an independent route.

Builds on: Hamilton's Equations · Phase Space · Partial Derivatives and the Chain Rule Leads to: Algebra of Poisson Brackets · Fundamental Poisson Brackets · Poisson-Bracket Formulation of Dynamics · Constants of Motion via Poisson Brackets · Poisson Brackets as a Test for Canonicity · Symplectic Condition

Is the value of a Poisson bracket something you could measure, or is it pure bookkeeping?

Measurable, very concretely. $\{f,H\}$ is the rate of change of $f$ — worked example 1's $-80\ \mathrm{J}$ is the slope a plot of $z(t)p(t)$ would show at that instant, and example 2's $-1.20\ \mathrm{N\,m}$ is a torque you could read off a torsion sensor. More generally $\{f,g\}$ is the answer to an experiment: "nudge the system along the flow generated by $g$ by a small parameter $\epsilon$ ; how much does $f$ shift, per unit $\epsilon$ ?" For $g=p$ the nudge is a spatial displacement, for $g=L_z$ a small rotation. Any bracket is a susceptibility of one observable to a transformation generated by another.

Why does the definition involve only first derivatives? Where did the second-order information go?

Because the bracket answers a first-order question — instantaneous rates along flows — and first-order questions need first derivatives. But there is a deeper design win: since $\{f,g\}$ involves only first derivatives, the result is again a legitimate phase-space function, which can itself be bracketed. That closure is what makes the observables an algebra and lets structures like $\{L_x,L_y\}=L_z$ exist. A "bracket" with second derivatives would produce objects outside the original space and the algebraic tower would collapse at the first floor.

How literally should I take "g generates a flow"? What flow, exactly?

Completely literally: solve the differential equations $\dot q_i=\partial g/\partial p_i$ , $\dot p_i=-\partial g/\partial q_i$ — Hamilton's equations with $g$ impersonating the Hamiltonian — and you get a one-parameter family of canonical maps, $g$ 's flow. For $g=p$ : $\dot q=1$ , $\dot p=0$ — rigid translation. For $g=L_z$ : rotation of both position and momentum about the $z$ -axis. For $g=H$ : the actual motion. The formal statement, via the type-2 machinery, is that $F_2=q_iP_i+\epsilon g$ generates the infinitesimal version; the bracket then reads out any observable's response, $\delta f=\epsilon\{f,g\}$ .

What exactly happens to brackets in quantum mechanics — and what breaks?

Dirac's rule replaces $\{f,g\}$ by $[\hat f,\hat g]/i\hbar$ : the fundamental bracket becomes $[\hat q,\hat p]=i\hbar$ , evolution (5) becomes Heisenberg's equation, and conserved quantities are operators commuting with $\hat H$ — the entire grammar transfers. What breaks is uniqueness beyond quadratic observables: ordering ambiguities ( $qp$ vs $\tfrac12(\hat q\hat p+\hat p\hat q)$ ) mean no consistent map makes all classical brackets go to commutators exactly (the Groenewold–van Hove theorem). The honest statement: commutators equal quantized brackets up to $O(\hbar^2)$ corrections, organized by the Moyal bracket. Classical mechanics is the $\hbar\to0$ skeleton, not a subset.

Can I compute brackets in coordinates that are not canonical — spherical velocity components, say?

Yes, but the formula changes. Definition (1) presumes canonical pairs; in arbitrary variables $y^a$ the bracket becomes $\{f,g\}=\partial_af\,\Pi^{ab}\,\partial_bg$ with a position-dependent antisymmetric tensor $\Pi$ (the pushforward of $J$ ). This is routine in advanced practice — rigid-body dynamics in terms of body-frame angular momenta uses $\{L_i,L_j\}=-\epsilon_{ijk}L_k$ with no canonical pairs in sight, and fluid/plasma brackets are of this noncanonical type, often with degenerate directions (Casimirs). The fundamental-table form of this page is the Darboux-chart special case; the geometry underneath is chart-free.

Is the minus sign in the definition just convention, or does it carry physics?

The relative minus between the two products is physics: it is the antisymmetry of $J$ , without which the bracket would be a symmetric object with no flow interpretation, no $\{f,f\}=0$ , no conservation-from-symmetry structure. The overall sign is convention (see the warning box) — flipping it consistently flips every formula but no physics. A useful check on any convention: $\{q,p\}$ should be $+1$ and a free particle should drift forward, $\{x,p^2/2m\}=p/m>0$ for $p>0$ . If both come out right, your signs are coherent.

Why is bracket invariance such a big deal? Coordinates were always arbitrary.

Because most formulas are not chart-invariant — kinetic energy looks different in every chart, and $\partial f/\partial q$ has no chart-free meaning at all. The bracket is special: it is built from exactly the structure ( $J$ ) that canonical charts are defined to preserve, so bracket statements are automatically about the physics rather than the description. Practically this buys two things: conservation laws proved in one chart hold in all charts; and the invariance can be run backwards as a diagnostic — if $\{Q,P\}\neq1$ , your new variables are not canonical, and the number tells you the area-distortion factor. That reversal is the canonicity test of the companion article.

What is the relation between the Poisson bracket and the commutator of vector fields from differential geometry?

Tight and exact: mapping each function $f$ to its Hamiltonian vector field $X_f$ (the flow generator above) is a homomorphism — $X_{\{f,g\}}=-[X_f,X_g]$ (sign per convention). So the bracket of functions mirrors the Lie bracket of their flows: two observables with vanishing Poisson bracket generate commuting flows (rotating then translating along the symmetry axis equals translating then rotating), which is why involutive constants of motion let tori and action-angle variables exist. The kernel of the map is the constants — adding a constant to $f$ changes no flow — which is the classical shadow of quantum phase freedom.