Canonical Transformations and Generating Functions

Canonical transformations are the cornerstone of advanced Hamiltonian mechanics, enabling you to recast complex dynamical problems into simpler forms. By preserving the symplectic structure of phase space, these coordinate changes reveal hidden conserved quantities and facilitate the solution of integrable systems, from oscillators to planetary orbits, bridging classical dynamics to quantum and statistical physics.

Canonical Transformations as Symplectomorphisms

In Hamiltonian mechanics, canonical transformations are coordinate changes from old variables $(q,p)$ to new ones $(Q,P)$ that leave the form of Hamilton's equations invariant. This means if the old coordinates obey $\dot{q}=\partial H/\partial p$ and $\dot{p}=-\partial H/\partial q$, then the new coordinates satisfy $\dot{Q}=\partial K/\partial P$ and $\dot{P}=-\partial K/\partial Q$ for some new Hamiltonian $K$. Geometrically, these transformations are symplectomorphisms—smooth maps that preserve the symplectic two-form $\omega =d{p}_i\wedge d{q}_i$ (summed over degrees of freedom). The symplectic two-form encodes the phase space structure, and preserving it ensures that Poisson brackets, such as $\{q,p\}=1$, remain unchanged, i.e., $\{Q,P\}=1$. This invariance is crucial because it maintains the canonical structure regardless of coordinates, allowing you to analyze dynamics in frames where conservation laws become obvious. For graduate-level physics, viewing canonical transformations as symplectomorphisms connects classical mechanics to differential geometry, providing a powerful language for global analysis of dynamical systems.

Classification of Generating Functions

Instead of guessing transformations, you can systematically generate them using generating functions. These are scalar functions of mixed old and new coordinates whose partial derivatives define the transformation equations. They are classified into four basic types based on which variables they depend on: $F_1(q,Q,t)$, $F_2(q,P,t)$, $F_3(p,Q,t)$, and $F_4(p,P,t)$. Each type is applicable depending on which set of coordinates is independent in a given problem. For instance, $F_2(q,P,t)$ is often used because it naturally leads to transformations where the new momentum $P$ is a constant of motion, such as in action-angle variables. The classification ensures completeness—any canonical transformation can be represented by one of these types, provided the Hessian matrix of second derivatives is non-singular, guaranteeing invertibility. Choosing the right generating function is a strategic step in simplifying Hamiltonians, as it allows you to tailor coordinates to the symmetries of the system.

Derivation of Transformation Equations

The transformation equations are derived by requiring that the variational principle or Hamilton's equations hold in both coordinate systems. Consider a generating function $F_1(q,Q,t)$. Its total differential is $d{F}_1=(\partial F_1/\partial q)dq+(\partial F_1/\partial Q)dQ+(\partial F_1/\partial t)dt$. In Hamiltonian mechanics, the differential form $pdq-PdQ+(K-H)dt$ must be an exact differential for the transformation to be canonical. Setting this equal to $d{F}_1$, we match coefficients to obtain:

$$p=\frac{\partial F_1}{\partial q},P=-\frac{\partial F_1}{\partial Q},K=H+\frac{\partial F_1}{\partial t}.$$

For a type $F_2(q,P,t)$, we use a Legendre transform: $F_2(q,P,t)=F_1(q,Q,t)+QP$, leading to:

$$p=\frac{\partial F_2}{\partial q},Q=\frac{\partial F_2}{\partial P},K=H+\frac{\partial F_2}{\partial t}.$$

Similar equations hold for $F_3$ and $F_4$. This step-by-step derivation shows how generating functions encode the mapping between coordinates while updating the Hamiltonian if time-dependent. You must always ensure the transformation is invertible by checking that determinants like ${\partial }^2{F}_2/\partial q\partial P$ are non-zero.

Applications to Integrable Systems

Canonical transformations excel at simplifying Hamiltonians for integrable systems, where actions become constants. For the harmonic oscillator, with Hamiltonian $H=(p^2/m+m{\omega }^2{q}^2)/2$, we seek action-angle variables $(J,\theta )$ where $H$ depends only on $J$. Use a generating function $F_1(q,\theta )=(m\omega q^2/2)\cot \theta$. From the equations, we compute $p=\partial F_1/\partial q=m\omega q\cot \theta$ and $J=-\partial F_1/\partial \theta =(m\omega q^2/2){\csc }^2\theta$. Inverting these gives $q=\sqrt{2J/(m\omega )}\sin \theta$ and $p=\sqrt{2m\omega J}\cos \theta$, yielding $H=\omega J$. Here, $J$ is constant and $\theta =\omega t+\text{const.}$, trivially integrating the motion.

For the Kepler problem (planetary motion), the Hamiltonian in polar coordinates is $H=p_r^2/(2m)+L^2/(2m{r}^2)-k/r$, with $L$ the angular momentum. This system is integrable with three conserved quantities. Applying a canonical transformation to Delaunay variables—generated by a type $F_2$—yields action variables $J_r,J_{\phi },J_{\psi }$ corresponding to radial motion, angular momentum, and orientation. The new Hamiltonian becomes $H=-m{k}^2/(2(J_r+J_{\phi }{)}^2)$, independent of angles, so actions are constants and angles evolve linearly. This transformation reduces the complex two-body motion to simple integrals, highlighting the power of canonical methods in celestial mechanics.

Common Pitfalls

When working with canonical transformations, several errors can undermine your analysis. First, misapplying generating function types leads to incorrect equations; for example, using $F_1$ relations for an $F_2$ function. Always verify which variables are independent: if you have $F_2(q,P,t)$, the derived equations are $p=\partial F_2/\partial q$ and $Q=\partial F_2/\partial P$, not $P=-\partial F_2/\partial Q$. Second, ignoring time dependence in the Hamiltonian update is common; if $F$ depends explicitly on time, the new Hamiltonian is $K=H+\partial F/\partial t$, not just $H$. Omitting $\partial F/\partial t$ can introduce spurious dynamics. Third, overlooking invertibility conditions can render transformations singular. For instance, in the harmonic oscillator example, the generating function uses $\cot \theta$, which is singular at $\theta =0$; you must restrict domains to ensure one-to-one mapping. To avoid these, consistently check derivations with test cases and validate Poisson brackets after transformation.

Summary

• Canonical transformations are symplectomorphisms that preserve Hamilton's equations by maintaining the symplectic structure $dp\wedge dq$, ensuring Poisson bracket invariance.
• Generating functions are classified into four types ($F_1,F_2,F_3,F_4$) based on variable dependence, providing a systematic method to derive transformation equations via partial derivatives.
• Transformation equations, such as $p=\partial F/\partial q$ and $Q=\partial F/\partial P$, are derived from equating differential forms, with the Hamiltonian updating to $K=H+\partial F/\partial t$ for time-dependent cases.
• Applications to integrable systems like the harmonic oscillator and Kepler problem demonstrate how canonical transformations simplify Hamiltonians into action-angle variables, revealing constants of motion and enabling straightforward integration.