Symplectic/Canonical Embeddings

Updated 16 April 2026

Symplectic/Canonical embeddings are injective maps that preserve the symplectic form, ensuring the Hamiltonian dynamics remain intact.
They exhibit rigidity through C0-closedness and shape invariants, which underpin obstruction theories like non-squeezing and capacity constraints.
Modern computations employ deep learning techniques to realize these embeddings, enabling efficient transformation and analysis of complex Hamiltonians.

A symplectic (or canonical) embedding is an injective, differentiable map between symplectic manifolds that exactly pulls back the symplectic form of the target to that of the source, or more generally, a map that transforms one canonical phase space structure to another while preserving the essential properties of Hamiltonian dynamics. The class includes classical symplectic diffeomorphisms, contact embeddings in odd dimensions, and deep learning–discovered symplectic conjugacies in infinite-dimensional or data-driven settings. These embeddings underpin rigidity phenomena in symplectic topology, obstruction theory for Hamiltonian dynamics, and modern machine-learning models for physical systems, as disclosed in recent developments in both the rigorous and applied literature.

1. Definitions and Characterizations

A smooth embedding $\varphi: (M^{2n},\omega_1) \hookrightarrow (N^{2n},\omega_2)$ is symplectic if $\varphi^*\omega_2 = \omega_1$ on $M$ . For canonical transformations in Hamiltonian mechanics, the condition is equivalently the preservation of the standard symplectic form: in local canonical coordinates, the Jacobian $M = D\varphi$ satisfies $M J M^T = J$ , with $J$ the standard symplectic matrix.

Beyond exactness, $\epsilon$ -symplectic embeddings $\varphi$ are defined by $\|\varphi^* \omega_2 - \omega_1\| < \epsilon$ for some fixed Riemannian metric, and their $C^0$ -limits are quantitatively close to genuine symplectic embeddings as $\varphi^*\omega_2 = \omega_1$ 0 (Müller, 2018). For contact geometry, the appropriate notion is a coisotropic embedding with respect to a contact structure, characterized by its preservation of pre-Lagrangian submanifolds.

Symplectic/Canonical embeddings are fundamentally distinguished from smooth or volume-preserving embeddings by their rigidity and the presence of global and local obstructions.

2. $\varphi^*\omega_2 = \omega_1$ 1-Rigidity and Shape Invariants

A pivotal result is the $\varphi^*\omega_2 = \omega_1$ 2-characterization of symplectic embeddings via the shape invariant, which encodes the set of possible period classes for Lagrangian embeddings up to translation by closed forms. Precisely, a smooth embedding $\varphi^*\omega_2 = \omega_1$ 3 is symplectic if and only if, for every exact open set $\varphi^*\omega_2 = \omega_1$ 4 and for every closed $\varphi^*\omega_2 = \omega_1$ 5-manifold $\varphi^*\omega_2 = \omega_1$ 6, the $\varphi^*\omega_2 = \omega_1$ 7-shape is mapped into the $\varphi^*\omega_2 = \omega_1$ 8-shape under $\varphi^*\omega_2 = \omega_1$ 9, that is,

$M$ 0

for every open $M$ 1 (Müller, 2016). The shape invariant is open with respect to the Weinstein neighborhood topology, closely related to the potential periods of Lagrangian submanifolds and displacement energy lower bounds.

A key consequence is the $M$ 2-rigidity theorem: the uniform $M$ 3-limit of symplectic embeddings is again symplectic, establishing the $M$ 4-closedness of $M$ 5 in $M$ 6. The shape-based characterization applies also in the contact case, using coisotropic (pre-Lagrangian) embeddings and pre-images under symplectization; the corresponding "contact shape" provides an analogous rigidity theorem for contact embeddings.

3. Advanced Obstruction Theory and Symplectic Packing

Symplectic embeddings are subject to both global (volume, cohomology) and fine-scale (capacity-type) obstructions. In low dimensions, Embedded Contact Homology (ECH) capacities provide a complete sequence of numerical invariants controlling when a symplectic embedding exists between four-dimensional toric domains; for example, embedding a concave toric domain into a convex one is possible if and only if all ECH capacity inequalities are satisfied (Cristofaro-Gardiner, 2014).

For higher-dimensional manifolds, the Gromov non-squeezing theorem and its multiball extensions provide the prototypical capacity-type obstructions. The work of Opshtein established that any closed $M$ 7-manifold can be fully packed (up to arbitrarily small leftover) by finitely many ellipsoids via singular polarizations, generalizing Donaldson's method to the irrational case (Opshtein, 2010).

Emergent theories use Newton–Okounkov bodies to extend ball-packing and ellipsoid embedding obstructions to non-toric rational surfaces, and algebraic capacities arising from the nef cone intersect with ECH capacities in dictating embedding feasibility (Chaidez et al., 2022).

4. Symplectic/Canonical Embeddings in Computational and Data-Driven Settings

Canonical transformations have been subsumed into deep learning architectures for data-driven modeling of Hamiltonian systems. Neural canonical transformations parameterized as symplectic networks (e.g., by enforcing $M$ 8 at each layer) are trained to map complicated Hamiltonians to decoupled latent forms, preserving phase-space measures and structural integrals. This paradigm appears in learning optimal collective coordinates for molecular dynamics, nonlinear system identification, and enabling efficient generative models for physical and image data (Li et al., 2019).

More broadly, deep learning-based approaches search for high-dimensional symplectic embeddings (including into overparameterized or lifted spaces) that preserve canonical Hamiltonian structure, enforce stability (e.g., via sum-of-squares Hamiltonian learning), and allow for both linear (Koopman-inspired) and cubic (weakly nonlinear) embeddings. These are robustly validated by machine-precision satisfaction of the symplecticity constraint and global Lyapunov stability (Goyal et al., 2023).

5. Hierarchies, Complete Embeddings, and Functoriality

A hierarchy arises between exact, tame, Kähler-type, and complete embeddings, with increasing structural requirements:

Exact symplectic embeddings require strict preservation of $M$ 9.
Tame/Kähler-type packings require holomorphicity with respect to compatible or tamed almost complex structures, with rigidity results showing uniqueness up to homology-trivial symplectomorphisms for torus and K3 targets with irrational symplectic forms (Entov et al., 2022).
Complete embeddings (for Liouville or geometrically bounded domains) admit functoriality of relative symplectic cohomology, encoding locality for Floer theoretical invariants, and mirror symmetry wall-crossing phenomena (Groman et al., 2021).

In the context of mirror symmetry and cluster varieties, canonical (complete) embeddings implement integral-affine charts on the base of Lagrangian torus fibrations, providing explicit global and local models for symplectic topology and the computation of Floer-theoretic invariants.

6. Contact and Infinite-Dimensional Extensions

Canonical embeddings generalize further in the context of contact geometry and infinite-dimensional symplectic and Stiefel/Grassmannian settings. Nonlinear analogues of Stiefel manifolds—spaces of "weighted embeddings" into a contact manifold—carry canonical symplectic forms commuting with the contact diffeomorphism group and function as the domains of symplectic reduction, giving rise to ‘dual pairs’ and providing a geometric description of certain coadjoint orbits (Haller et al., 2019).

Moreover, tables of symplectic capacities and explicit combinatorial weight expansions (e.g., ECH capacity weight-sequences, Newton–Okounkov weights) provide algorithmic criteria for the existence and classification of canonical embeddings in various categories.

7. Quantitative and Topological Flexibility

The quantitative behavior of near-symplectic (or $M = D\varphi$ 0-symplectic) embeddings ensures a stable control both for numerical and topologically flexible applications. $M = D\varphi$ 1-rigidity extends to $M = D\varphi$ 2-symplectic embeddings, showing that if a sequence of $M = D\varphi$ 3-symplectic embeddings converges uniformly, the limit is $M = D\varphi$ 4-symplectic with $M = D\varphi$ 5 (Müller, 2018). This feature supports robustness for symplectic numerical integrators and approximate symplectic structures in computational physics and quantum computation.

Furthermore, universal constructions using infinite-dimensional projective limits of ambient manifolds allow parametric, functorial realization of all closed $M = D\varphi$ 6-forms as pullbacks via canonical embeddings, leading to weak Serre fibration-like behavior for the embedding spaces (Araujo et al., 2014).

The study of symplectic/canonical embeddings thus lies at the interface of geometric analysis, obstruction theory, Floer-type invariants, and modern computational models, encoding both profound rigidity and considerable flexibility depending on the ambient structures and embedding category.