Symplectic Structure and Hamiltonian Normal Form

Author: Tianjiang Shuo
Website: https://cislunarspace.cn

Symplectic matrices and transformations

On the standard phase space $\mathbb{R}^{2n}$ with coordinates $\mathbf{z}=(\mathbf{q},\mathbf{p})$ define the block matrix $J=\begin{pmatrix}0&I\\-I&0\end{pmatrix}$ . A real $2n\times 2n$ matrix $M$ is symplectic if $M^{\!\top}JM=J$ ; the set of all such matrices forms the symplectic group $Sp(2n,\mathbb{R})$ . Symplectic matrices have determinant 1 and spectra symmetric under $\lambda\mapsto\bar\lambda$ , $\lambda\mapsto 1/\lambda$ (Meyer & Offin 2017, §2.4). A smooth coordinate change $\mathbf{z}=\mathbf{Z}(\boldsymbol{\zeta})$ is a symplectic (canonical) transformation if its Jacobian is everywhere symplectic; such transformations preserve the form of Hamilton's equations.

The state-transition matrix of any Hamiltonian flow is symplectic — this is why the monodromy matrix of a libration-point orbit has the saddle×center×center eigenvalue structure, and why Poincaré maps preserve phase-space volume. Symplectic geometry is the differential-geometric abstraction: a symplectic manifold carries a closed, non-degenerate 2-form $\omega$ ; Darboux's theorem guarantees that locally all symplectic manifolds look like $(\mathbb{R}^{2n},\omega_0)$ .

Poisson bracket

For smooth functions $F,G$ on phase space, the Poisson bracket is

\{F,G\}=\nabla F^{\!\top}J\,\nabla G=\frac{\partial F}{\partial\mathbf{q}^{\!\top}}\frac{\partial G}{\partial\mathbf{p}}-\frac{\partial F}{\partial\mathbf{p}^{\!\top}}\frac{\partial G}{\partial\mathbf{q}}.

It is bilinear, antisymmetric, and satisfies the Jacobi identity, turning smooth functions into a Lie algebra. Hamilton's equations read $\dot F=\{F,H\}$ . A function $F$ is conserved iff $\{F,H\}=0$ ; $n$ independent integrals in involution (mutual Poisson brackets zero) make the system integrable. The Poisson bracket is invariant under symplectic changes of variables — this is the algebraic content of Theorem 2.6.3 of Meyer & Offin (2017), and the reason canonical transformations are the natural changes of variables in Hamiltonian mechanics.

Generating functions and Lie transforms

A canonical transformation can be encoded by a generating function. In the Lie-series approach (Hori 1966; Deprit 1969; Meyer), an autonomous Hamiltonian $G$ generates a one-parameter family of canonical transformations $\boldsymbol{\zeta}\mapsto\mathbf{Z}_t(\boldsymbol{\zeta})$ whose action on any function $F$ is $F\mapsto F+\{G,F\}+\tfrac12\{G,\{G,F\}\}+\cdots$ — a Lie transform. Choosing $G$ to cancel unwanted terms in $H$ order by order yields the normal form. The Hori/Deprit method provides a recursive, computer-algebra-friendly implementation widely used in celestial mechanics.

Normal-form procedure and the homological equation

Consider a Hamiltonian near an equilibrium (e.g. a collinear libration point) expanded as $H=H_2+\sum_{k\geq 3}H_k$ with $H_k$ homogeneous of degree $k$ . A normal form seeks a canonical transformation $\boldsymbol{\zeta}\mapsto\mathbf{Z}(\boldsymbol{\zeta})$ given as the time-one flow of a generating Hamiltonian $G=\sum_{k\geq 3}G_k$ , so that in the new variables the unwanted terms vanish order by order (Gómez et al. 2001, vol. III). At degree $k$ the determining equation is the homological equation

\widetilde H_k = H_k + \{G_k,H_2\} + (\text{lower-order knowns}),

whose unknowns are $G_k$ and the new coefficient polynomial $\widetilde H_k$ . Each monomial of $H_k$ contributes a denominator of the form $\mathbf{k}\cdot\boldsymbol{\omega}$ (the frequency dot-product) when solved for $G_k$ . If $\mathbf{k}\cdot\boldsymbol{\omega}$ is small but non-zero, the corresponding coefficient of $G_k$ is huge — the small-divisor problem — and the formal series may diverge. At collinear points the hyperbolic frequency $\lambda$ bounds the denominators away from zero for non-resonant terms, so reduction to any finite order is well-defined.

Birkhoff and Birkhoff–Gustavson normal forms

The Birkhoff normal form (BNF) eliminates all non-resonant monomials, leaving only terms in involution with $H_2$ ; in the non-resonant case the normalized Hamiltonian depends only on the actions, making the truncated system integrable (Birkhoff 1927; Meyer & Offin 2017, Ch.10). The Birkhoff–Gustavson normal form handles resonant cases by retaining resonant monomials — appropriate for the spatial CR3BP where the two center frequencies are near the 1:1 resonance (giving rise to halo orbits) and higher-order resonances appear near order 57 for the Hill case (Gómez et al. 2001, vol. III).

Partial normal form, reduction, and Moser's theorem

A partial normal form (PNF) cancels only the unstable (hyperbolic) terms, leaving the center-manifold dynamics intact. Combined with reduction to the center manifold $W^c$ , this yields a 4-D (or 2-D planar) conservative Hamiltonian that captures Lissajous, halo, and quasi-halo families near $L_{1,2}$ without the hyperbolic directions (Gómez et al. 2001; Jorba & Masdemont 1999). Normal-form reduction is the same idea applied to the Poincaré map: simplify the symplectic map to read off stability and bifurcation structure.

Moser's theorem (Moser 1958) gives sufficient conditions under which the Birkhoff normal form converges near an elliptic equilibrium — the formal series then describes a true invariant curve, justifying the use of normal forms for stability claims and parameterization of invariant manifolds near libration points.

Application notes

Normal-form computations at $L_1/L_2$ to order ~15–35 give accurate semi-analytical approximations of halo and Lissajous orbits used as initial guesses for differential correction.
Small divisors set the practical order limit; near low-order resonances one must switch to resonant normal forms.
Computer-algebra implementations of the Hori/Deprit scheme (symbolic manipulators) are the workhorse tools; partial normal forms are preferred when only the center dynamics is needed.

References

Meyer, K. R., & Offin, D. C. (2017). Introduction to Hamiltonian Dynamical Systems and the N-Body Problem, 3rd ed., Ch. 2, 7, 10.
Gómez, G., Llibre, J., & Martínez, R. (2001). Dynamics and Mission Design near Libration Points, vol. III — Advanced Methods for Collinear Points.
Moser, J. (1958). New aspects in the theory of stability of Hamiltonian systems. Comm. Pure Appl. Math., 11, 81–114.
Hori, G.-I. (1966). Theory of general perturbations. PASJ, 18, 287–296.
Deprit, A. (1969). Canonical transformations depending on a small parameter. CeMec, 1, 12–30.
Birkhoff, G. D. (1927). Dynamical Systems.
Celletti, A., et al. (2024). The dynamics around the collinear points of the elliptic three-body problem: a normal form approach.