Skip to content

Poisson Brackets and Canonical Transformations

CSIR NET advanced mechanics: Poisson bracket algebra, canonicity tests, symplectic condition, four generating functions and infinitesimal transformations.

  • 21 min read
  • Updated
On this page · 16 sections
  1. Prerequisites
  2. The Hook: Turning an Oscillator into a Rotor
  3. Poisson Brackets: the Full Algebra
  4. Symplectic Notation
  5. Canonical Transformations
  6. Generating Functions
  7. Infinitesimal Canonical Transformations
  8. Lagrange Brackets
  9. A Time-Dependent Example: the Damped Oscillator
  10. Memory Tricks
  11. Don’t Confuse
  12. Common Mistakes
  13. Practice
  14. Quick Revision Table
  15. Key Takeaways
  16. What’s Next

Prerequisites

The Hook: Turning an Oscillator into a Rotor

Poisson Brackets: the Full Algebra

For \( n \) degrees of freedom and functions \( f, g \) of \( (q, p, t) \), with the sign convention of Goldstein and of the core page,

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

Some books (Landau–Lifshitz) define the bracket with the opposite sign. Check the fundamental bracket a question uses: in this convention \( \{q, p\} = +1 \).

Properties

PropertyStatement
Antisymmetry\( \{f, g\} = -\{g, f\} \), so \( \{f, f\} = 0 \)
Bilinearity\( \{af + bg, h\} = a\{f, h\} + b\{g, h\} \) for constants \( a, b \)
Leibniz (product) rule\( \{fg, h\} = f\{g, h\} + \{f, h\}g \)
Chain rule\( \{f, \phi(g)\} = \phi'(g)\{f, g\} \)
Jacobi identity\( \{f, \{g, h\}\} + \{g, \{h, f\}\} + \{h, \{f, g\}\} = 0 \)
Constants\( \{c, f\} = 0 \)

Fundamental brackets and shortcuts

$$\{q_i, q_j\} = 0, \qquad \{p_i, p_j\} = 0, \qquad \{q_i, p_j\} = \delta_{ij}$$

Two shortcuts follow from the definition and save most of the work:

$$\{q_i, f\} = \frac{\partial f}{\partial p_i}, \qquad \{p_i, f\} = -\frac{\partial f}{\partial q_i}$$

So \( \{x, p_x^2\} = 2p_x \) and \( \{p_x, V(x)\} = -V'(x) \) without writing a sum.

Angular momentum

With \( \mathbf{L} = \mathbf{r}\times\mathbf{p} \):

$$\{L_i, L_j\} = \varepsilon_{ijk}L_k, \qquad \{L^2, L_i\} = 0, \qquad \{x_i, L_j\} = \varepsilon_{ijk}x_k, \qquad \{p_i, L_j\} = \varepsilon_{ijk}p_k$$

Any scalar built from \( \mathbf{r} \) and \( \mathbf{p} \) (such as \( r^2 \), \( p^2 \), \( \mathbf{r}\cdot\mathbf{p} \)) has zero bracket with every \( L_i \). A vector \( \mathbf{A} \) built from them satisfies \( \{A_i, L_j\} = \varepsilon_{ijk}A_k \). These two facts settle most angular-momentum bracket questions in one line.

Dynamics and conservation

$$\boxed{\frac{df}{dt} = \{f, H\} + \frac{\partial f}{\partial t}}$$

Hence:

  • Hamilton’s equations: \( \dot q_i = \{q_i, H\} \), \( \dot p_i = \{p_i, H\} \).
  • \( f \) with no explicit time dependence is conserved if and only if \( \{f, H\} = 0 \).
  • Poisson’s theorem: if \( f \) and \( g \) are constants of motion, so is \( \{f, g\} \). (Proof: Jacobi identity plus the explicit-time terms.) It can give something new, or a function of known integrals, or zero.
  • The formal solution is a series: \( f(t) = f + t\{f, H\} + \tfrac{t^2}{2!}\{\{f, H\}, H\} + \cdots \), evaluated at \( t = 0 \).

The quantum correspondence

Dirac’s rule \( \{f, g\} \to [\hat f, \hat g]/i\hbar \) carries the whole algebra into quantum mechanics: \( [\hat x, \hat p] = i\hbar \), \( [\hat L_x, \hat L_y] = i\hbar\hat L_z \), and \( d\langle A\rangle/dt = \langle[\hat A, \hat H]\rangle/i\hbar \). Canonical transformations correspond to unitary transformations, and the generator \( G \) of an infinitesimal canonical transformation (below) becomes the Hermitian generator in \( \hat U = 1 - i\epsilon\hat G/\hbar \).

Symplectic Notation

Collect the phase-space coordinates into \( \boldsymbol\eta = (q_1, \ldots, q_n, p_1, \ldots, p_n) \) and define the \( 2n\times2n \) matrix

$$\Omega = \begin{pmatrix}0 & I_n\\ -I_n & 0\end{pmatrix}, \qquad \Omega^T = -\Omega = \Omega^{-1}$$

Then Hamilton’s equations and the bracket take compact forms:

$$\dot{\boldsymbol\eta} = \Omega\,\frac{\partial H}{\partial\boldsymbol\eta}, \qquad \{f, g\} = \left(\frac{\partial f}{\partial\boldsymbol\eta}\right)^T\Omega\,\frac{\partial g}{\partial\boldsymbol\eta}, \qquad \{\eta_a, \eta_b\} = \Omega_{ab}$$

Canonical Transformations

A transformation \( (q, p) \to (Q, P) \), possibly time-dependent, is canonical if for every Hamiltonian \( H(q, p, t) \) there is a function \( K(Q, P, t) \) such that

$$\dot Q_i = \frac{\partial K}{\partial P_i}, \qquad \dot P_i = -\frac{\partial K}{\partial Q_i}$$

Four equivalent tests

Use whichever is fastest for the question.

TestConditionBest for
Fundamental brackets\( \{Q_i, Q_j\} = 0 \), \( \{P_i, P_j\} = 0 \), \( \{Q_i, P_j\} = \delta_{ij} \), computed in \( (q, p) \)Explicit \( Q(q, p) \), \( P(q, p) \) with parameters to fix
Symplectic conditionJacobian \( M = \partial(Q, P)/\partial(q, p) \) satisfies \( M\Omega M^T = \Omega \)Linear transformations
One degree of freedom\( \{Q, P\} = \dfrac{\partial(Q, P)}{\partial(q, p)} = 1 \) (the Jacobian determinant)Most CSIR NET questions
Differential form\( \sum p_i\,dq_i - \sum P_i\,dQ_i \) is an exact differential (at fixed \( t \))Finding the generating function

For one degree of freedom, \( \{Q, P\} = 1 \) is the whole story: it says the transformation preserves area in the \( (q, p) \) plane. For the linear map \( Q = aq + bp \), \( P = cq + dp \), the condition is simply \( ad - bc = 1 \).

Consequences of canonicity

  • Poisson brackets are invariant: \( \{f, g\}_{q,p} = \{f, g\}_{Q,P} \). That is why the subscript is usually dropped.
  • Phase-space volume is invariant: the Jacobian determinant of a canonical transformation is 1. Liouville’s theorem follows, because time evolution is itself a canonical transformation.
  • Composition and inverse of canonical transformations are canonical: they form a group.

Generating Functions

The variational principle for both sets of variables differs at most by a total time derivative:

$$\sum_i p_i\dot q_i - H = \sum_i P_i\dot Q_i - K + \frac{dF}{dt}$$

\( F \) is the generating function. Choosing it as a function of one old and one new variable gives four types.

TypeVariablesOld variable fromNew variable from
\( F_1(q, Q, t) \)\( q, Q \)\( p_i = \dfrac{\partial F_1}{\partial q_i} \)\( P_i = -\dfrac{\partial F_1}{\partial Q_i} \)
\( F_2(q, P, t) \)\( q, P \)\( p_i = \dfrac{\partial F_2}{\partial q_i} \)\( Q_i = \dfrac{\partial F_2}{\partial P_i} \)
\( F_3(p, Q, t) \)\( p, Q \)\( q_i = -\dfrac{\partial F_3}{\partial p_i} \)\( P_i = -\dfrac{\partial F_3}{\partial Q_i} \)
\( F_4(p, P, t) \)\( p, P \)\( q_i = -\dfrac{\partial F_4}{\partial p_i} \)\( Q_i = \dfrac{\partial F_4}{\partial P_i} \)

In every case

$$\boxed{K = H + \frac{\partial F}{\partial t}}$$

so a time-independent generating function leaves \( K = H \) (expressed in the new variables). The types are related by Legendre transformations, for example \( F_2 = F_1 + \sum Q_iP_i \).

Standard generating functions

Generating functionTransformationMeaning
\( F_2 = \sum q_iP_i \)\( Q = q \), \( P = p \)Identity
\( F_2 = \sum f_i(q)P_i \)\( Q_i = f_i(q) \), momenta transform accordinglyPoint transformation (every coordinate change is canonical once momenta are carried along)
\( F_1 = \sum q_iQ_i \)\( Q_i = p_i \), \( P_i = -q_i \)Exchange: coordinates and momenta are on an equal footing
\( F_2 = \sum(q_i + a_i)P_i \)\( Q = q + a \), \( P = p \)Translation
\( F_1 = \tfrac{1}{2}m\omega q^2\cot Q \)\( q = \sqrt{2P/m\omega}\sin Q \), \( p = \sqrt{2m\omega P}\cos Q \)Oscillator to angle-type variables; \( K = \omega P \)
\( F_2 = qPe^{\gamma t} \)\( Q = qe^{\gamma t} \), \( P = pe^{-\gamma t} \)Time-dependent rescaling; \( K = H + \gamma QP \)

Finding a generating function from a given transformation

  1. Check canonicity first (\( \{Q, P\} = 1 \)).
  2. Pick a type whose two variables are independent for this transformation (if \( Q = Q(q) \) alone, \( q \) and \( Q \) are not independent, so \( F_1 \) fails; use \( F_2 \) or \( F_3 \)).
  3. Express the two remaining variables in terms of the chosen pair and integrate the two partial-derivative equations, checking that they are consistent.

Infinitesimal Canonical Transformations

Take \( F_2 = \sum q_iP_i + \epsilon\,G(q, P) \) with \( \epsilon \) small. To first order in \( \epsilon \),

$$\delta q_i = \epsilon\frac{\partial G}{\partial p_i}, \qquad \delta p_i = -\epsilon\frac{\partial G}{\partial q_i}, \qquad \boxed{\delta f = \epsilon\{f, G\}}$$

\( G \) is the generator of the transformation.

Generator \( G \)Transformation generated
\( p_k \)Translation of \( q_k \): \( \delta q_k = \epsilon \)
\( \hat{\mathbf{n}}\cdot\mathbf{P} \) (total momentum)Translation of the whole system along \( \hat{\mathbf{n}} \)
\( L_z = xp_y - yp_x \)Rotation about \( z \): \( \delta x = -\epsilon y \), \( \delta y = \epsilon x \)
\( H \), with \( \epsilon = dt \)Time evolution: \( \delta f = dt\{f, H\} \)

The last line means the motion itself is a continuous canonical transformation, generated by the Hamiltonian. This is the deepest reason why Liouville’s theorem holds and why Poisson brackets computed at time \( t \) equal those at time 0.

The change of the Hamiltonian under the transformation generated by \( G \) is \( \delta H = \epsilon\{H, G\} \). So \( G \) leaves \( H \) invariant exactly when \( \{G, H\} = 0 \), which is exactly when \( G \) is conserved. This is Noether’s theorem in Hamiltonian form, developed in Symmetry, Invariance and Noether’s Theorem.

Lagrange Brackets

For completeness, the Lagrange bracket of two variables \( u, v \) is

$$[u, v] = \sum_i\left(\frac{\partial q_i}{\partial u}\frac{\partial p_i}{\partial v} - \frac{\partial p_i}{\partial u}\frac{\partial q_i}{\partial v}\right)$$

It is also invariant under canonical transformations, with fundamental values \( [q_i, p_j] = \delta_{ij} \), \( [q_i, q_j] = [p_i, p_j] = 0 \). For a set of \( 2n \) independent functions \( u_1, \ldots, u_{2n} \) the two kinds of bracket are reciprocal: \( \sum_l[u_l, u_i]\{u_l, u_j\} = \delta_{ij} \). For a pair, \( [Q, P] = 1 \) if and only if \( \{Q, P\} = 1 \). In exams it appears only as an alternative canonicity test.

A Time-Dependent Example: the Damped Oscillator

The Hamiltonian

$$H = e^{-\gamma t}\frac{p^2}{2m} + e^{\gamma t}\frac{m\omega^2q^2}{2}$$

gives \( \dot q = e^{-\gamma t}p/m \), so \( p = me^{\gamma t}\dot q \), and \( \dot p = -e^{\gamma t}m\omega^2q \). Eliminating \( p \): \( \ddot q + \gamma\dot q + \omega^2q = 0 \), the damped oscillator (the Caldirola–Kanai Hamiltonian). \( H \) depends on time and is not conserved.

Apply \( F_2 = qPe^{\gamma t/2} \): \( Q = qe^{\gamma t/2} \), \( P = pe^{-\gamma t/2} \), and

$$K = H + \frac{\partial F_2}{\partial t} = \frac{P^2}{2m} + \frac{m\omega^2Q^2}{2} + \frac{\gamma}{2}QP$$

\( K \) has no explicit time dependence, so it is conserved. Back in the old variables, the conserved quantity is \( H + \tfrac{\gamma}{2}qp \). A canonical transformation found a hidden constant of motion.

Memory Tricks

Don’t Confuse

ThisvsThat
Poisson bracket \( \{q, p\} = 1 \) (Goldstein)Landau–Lifshitz sign: \( \{p, q\} = 1 \)
Canonical: \( \{Q, P\} = 1 \)Scale transformation: \( \{Q, P\} = \lambda\mu \), \( K = \lambda\mu H \)
\( F_1 \): \( P = -\partial F_1/\partial Q \)\( F_2 \): \( Q = +\partial F_2/\partial P \)
\( K = H \) for time-independent \( F \)\( K = H + \partial F/\partial t \) otherwise
Point transformation (\( Q = Q(q) \)): always canonical with suitable \( P \)\( F_1(q, Q) \) cannot generate it (\( q \) and \( Q \) not independent)
\( \{f, H\} = 0 \): conserved if \( f \) has no explicit \( t \)Explicitly time-dependent \( f \): need \( \{f, H\} + \partial f/\partial t = 0 \)

Common Mistakes

Practice

All questions below are original practice questions written for this page. They are not from any CSIR NET paper.

Level 1: Foundation

  1. Evaluate \( \{x, p_x^2 + p_y^2\} \).
  2. What transformation does \( F_2 = \sum_iq_iP_i \) generate?
  3. Evaluate \( \{L_x, L_y\} \).
Answers
  1. \( 2p_x \). \( \{x, f\} = \partial f/\partial p_x \).
  2. The identity: \( p_i = \partial F_2/\partial q_i = P_i \) and \( Q_i = \partial F_2/\partial P_i = q_i \).
  3. \( L_z \). From \( \{L_i, L_j\} = \varepsilon_{ijk}L_k \); directly, \( x p_y - yp_x \).

Level 2: Part B Style

Practice Q1. The transformation \( Q = q^\alpha e^{\beta p} \), \( P = q^\alpha e^{-\beta p} \) (for \( q > 0 \)) is canonical for:

(a) \( \alpha = 1 \), \( \beta = 1 \) (b) \( \alpha = 1/2 \), \( \beta = 1 \) (c) \( \alpha = 1/2 \), \( \beta = -1 \) (d) \( \alpha = 2 \), \( \beta = -1/4 \)

Answer: (c)
$$\{Q, P\} = \alpha q^{\alpha-1}e^{\beta p}\cdot(-\beta q^\alpha e^{-\beta p}) - \beta q^\alpha e^{\beta p}\cdot\alpha q^{\alpha-1}e^{-\beta p} = -2\alpha\beta\,q^{2\alpha-1}$$

This must equal 1 for all \( q \): \( 2\alpha - 1 = 0 \) and \( -2\alpha\beta = 1 \), so \( \alpha = 1/2 \), \( \beta = -1 \). Option (b) gives \( -1 \): area-preserving but orientation-reversing, which is not canonical. Option (d) makes the bracket \( q^3 \), not constant.

Practice Q2. For \( H = \dfrac{p^2}{2m} + \dfrac{1}{2}m\omega^2q^2 \) and the transformation generated by \( F_1 = \tfrac{1}{2}m\omega q^2\cot Q \), Hamilton’s equations in the new variables give:

(a) \( \dot Q = \omega \), \( \dot P = 0 \) (b) \( \dot Q = 0 \), \( \dot P = \omega \) (c) \( \dot Q = \omega P \), \( \dot P = 0 \) (d) \( \dot Q = \omega \), \( \dot P = -\omega Q \)

Answer: (a)

\( K = \omega P \). \( \dot Q = \partial K/\partial P = \omega \) and \( \dot P = -\partial K/\partial Q = 0 \). The new momentum is \( E/\omega \), constant, and the new coordinate is an angle advancing uniformly.

Practice Q3. Under the infinitesimal canonical transformation generated by \( G = xp_y - yp_x \), the changes in \( x \) and \( y \) are:

(a) \( \delta x = \epsilon y \), \( \delta y = -\epsilon x \) (b) \( \delta x = -\epsilon y \), \( \delta y = \epsilon x \) (c) \( \delta x = \epsilon p_y \), \( \delta y = -\epsilon p_x \) (d) \( \delta x = \delta y = 0 \)

Answer: (b)

\( \delta x = \epsilon\,\partial G/\partial p_x = -\epsilon y \) and \( \delta y = \epsilon\,\partial G/\partial p_y = \epsilon x \). This is an anticlockwise rotation by angle \( \epsilon \) about \( z \): \( G = L_z \) generates rotations.

Practice Q4. A generating function of type \( F_2(q, P) \) for the canonical transformation \( Q = q + aP^2 \), \( P = p \) is:

(a) \( qP + aP^2 \) (b) \( qP + aP^3/3 \) (c) \( qP - aP^3/3 \) (d) \( qP + aqP^2 \)

Answer: (b)

\( \partial F_2/\partial q = P = p \) and \( \partial F_2/\partial P = q + aP^2 = Q \). Option (d) gives \( p = P + aP^2 \), which is wrong.

Level 3: Part C Style

Practice Q5. Which of the following transformations are canonical, with the stated new Hamiltonian?

  1. \( Q = p \), \( P = -q \), with \( K(Q, P) = H(-P, Q) \).
  2. \( Q = 2q \), \( P = 2p \), with \( K = H \).
  3. \( Q = qe^{\gamma t} \), \( P = pe^{-\gamma t} \), with \( K = H + \gamma QP \).
  4. \( Q = 2q + p \), \( P = 3q + 2p \), with \( K = H \).

(a) 1 and 3 only (b) 1, 3 and 4 only (c) 2 and 4 only (d) 1, 2, 3 and 4

Answer: (b) 1, 3 and 4 only
  1. \( \{p, -q\} = -\{p, q\} = 1 \). Canonical, generated by \( F_1 = qQ \) (time-independent, so \( K \) is \( H \) with \( q = -P \), \( p = Q \)). True.
  2. \( \{2q, 2p\} = 4 \neq 1 \). Not canonical. False.
  3. \( \{qe^{\gamma t}, pe^{-\gamma t}\} = 1 \). Generated by \( F_2 = qPe^{\gamma t} \); \( \partial F_2/\partial t = \gamma qPe^{\gamma t} = \gamma QP \). True.
  4. Linear with \( a = 2 \), \( b = 1 \), \( c = 3 \), \( d = 2 \): \( ad - bc = 4 - 3 = 1 \). Canonical, time-independent, so \( K = H \). True.

Practice Q6. For the Hamiltonian \( H = e^{-\gamma t}\dfrac{p^2}{2m} + e^{\gamma t}\dfrac{m\omega^2q^2}{2} \), which of the following is a constant of motion?

(a) \( H \) (b) \( H + \gamma qp \) (c) \( H + \tfrac{\gamma}{2}qp \) (d) \( e^{\gamma t}H \)

Answer: (c)

The transformation \( Q = qe^{\gamma t/2} \), \( P = pe^{-\gamma t/2} \) (from \( F_2 = qPe^{\gamma t/2} \)) gives the time-independent \( K = P^2/2m + m\omega^2Q^2/2 + \tfrac{\gamma}{2}QP \), and \( QP = qp \).

Direct check: let \( C = H + \tfrac{\gamma}{2}qp \). Then \( \partial C/\partial t = -\gamma e^{-\gamma t}p^2/2m + \gamma e^{\gamma t}m\omega^2q^2/2 \). Also \( \{H, H\} = 0 \) and

$$\left\{\tfrac{\gamma}{2}qp, H\right\} = \tfrac{\gamma}{2}\left(q\{p, H\} + p\{q, H\}\right) = \tfrac{\gamma}{2}\left(-e^{\gamma t}m\omega^2q^2 + e^{-\gamma t}\frac{p^2}{m}\right)$$

The sum \( \{C, H\} + \partial C/\partial t = 0 \). Option (a) fails because \( H \) depends explicitly on time.

Practice Q7. With \( \mathbf{L} = \mathbf{r}\times\mathbf{p} \), which of the following brackets are correct?

  1. \( \{L_z, x\} = y \)
  2. \( \{L_z, p_x\} = p_y \)
  3. \( \{L_x, L_y\} = -L_z \)
  4. \( \{L^2, L_z\} = 0 \)

(a) 1, 2 and 4 only (b) 2 and 4 only (c) 1 and 3 only (d) 1, 2, 3 and 4

Answer: (a) 1, 2 and 4 only
  1. \( \{x, L_z\} = \varepsilon_{xzy}y = -y \), so \( \{L_z, x\} = y \). Directly: \( \{L_z, x\} = -\partial L_z/\partial p_x = -(-y) = y \). True.
  2. \( \{L_z, p_x\} = \partial L_z/\partial x = p_y \). True.
  3. \( \{L_x, L_y\} = +L_z \). False (sign).
  4. \( L^2 \) is a scalar. True.

Quick Revision Table

IdeaResult
Bracket\( \{f, g\} = \sum(\partial_qf\,\partial_pg - \partial_pf\,\partial_qg) \)
Shortcuts\( \{q_i, f\} = \partial f/\partial p_i \), \( \{p_i, f\} = -\partial f/\partial q_i \)
Time evolution\( \dot f = \{f, H\} + \partial_tf \)
Poisson’s theorem\( \{f, g\} \) of two constants is a constant
Angular momentum\( \{L_i, L_j\} = \varepsilon_{ijk}L_k \); scalars commute with \( \mathbf{L} \)
Canonical (1D)\( \{Q, P\} = 1 \); linear: \( ad - bc = 1 \)
Symplectic\( M\Omega M^T = \Omega \)
Generating functions\( F_1 \): \( p = \partial_qF \), \( P = -\partial_QF \); \( F_2 \): \( p = \partial_qF \), \( Q = \partial_PF \)
New Hamiltonian\( K = H + \partial F/\partial t \)
Exchange\( F_1 = qQ \): \( Q = p \), \( P = -q \)
Infinitesimal\( \delta f = \epsilon\{f, G\} \); \( H \) generates time evolution
Invariantsbrackets, phase-space volume

Key Takeaways

What’s Next