Symplectic Billiards Dynamics

• Symplectic billiards are dynamical systems where reflections follow a symplectic or area constraint instead of Euclidean equal-angle rules.
• They are modeled as exact symplectic twist maps generated by an area functional, integrating convex geometry, Aubry–Mather theory, and symplectic topology.
• Key features include periodic orbit rigidity, affine equivariance, and extensions to dissipative and higher-dimensional cases, driving ongoing research inquiries.

Symplectic billiards are billiard-type dynamical systems in which the reflection law is governed by a symplectic or area constraint rather than by Euclidean equal-angle reflection. For a strictly convex smooth planar domain $\Omega$, consecutive impacts $p_{k-1},p_k,p_{k+1}\in \partial\Omega$ satisfy that the line through $p_{k-1}$ and $p_{k+1}$ is parallel to the tangent line at $p_k$; in $\mathbb R^{2n}$ the tangent direction is replaced by the characteristic line $$\ker(\omega|_{T_{p_k}\partial\Omega})=\mathbb R JN_{p_k}$$ determined by the ambient symplectic form and the outward Euclidean normal (Sharipova, 2024). In the planar case the dynamics is naturally encoded as an exact symplectic twist map generated by an area functional, which places symplectic billiards at the intersection of convex geometry, Aubry–Mather theory, billiard dynamics, and symplectic topology (Baranzini et al., 8 Sep 2025).

1. Geometric definition and basic setup

Let $(\mathbb R^{2n},\omega)$ be the standard symplectic vector space, with

$\omega(u,v)=\sum_{i=1}^n (u_{q_i}v_{p_i}-u_{p_i}v_{q_i}) =\langle Ju,v\rangle, \qquad J=\begin{pmatrix}0&I_n\-I_n&0\end{pmatrix}.$

If $X\subset \mathbb R^{2n}$ is a strictly convex domain with smooth boundary and $p_{k-1},p_k,p_{k+1}\in \partial\Omega$0 is the outward unit normal at $p_{k-1},p_k,p_{k+1}\in \partial\Omega$1, then the restriction $p_{k-1},p_k,p_{k+1}\in \partial\Omega$2 has a one-dimensional kernel, and this characteristic direction is

$p_{k-1},p_k,p_{k+1}\in \partial\Omega$3

The higher-dimensional symplectic billiard reflection law is defined by requiring that for distinct $p_{k-1},p_k,p_{k+1}\in \partial\Omega$4, the line $p_{k-1},p_k,p_{k+1}\in \partial\Omega$5 be tangent to $p_{k-1},p_k,p_{k+1}\in \partial\Omega$6 at $p_{k-1},p_k,p_{k+1}\in \partial\Omega$7 in the characteristic direction $p_{k-1},p_k,p_{k+1}\in \partial\Omega$8; equivalently, the direction of the chord $p_{k-1},p_k,p_{k+1}\in \partial\Omega$9 lies in $p_{k-1}$0 (Sharipova, 2024).

In dimension two this reduces to a particularly transparent rule. If $p_{k-1}$1 is strictly convex with smooth boundary, then the characteristic line is simply the tangent line, and a triple $p_{k-1}$2 satisfies the reflection law precisely when

$p_{k-1}$3

Thus symplectic billiards are nonlocal in the sense that the condition is imposed on a triple of successive points rather than on a single incoming direction at a reflection point. The symplectic billiard map sends an oriented chord $p_{k-1}$4 to $p_{k-1}$5, where $p_{k-1}$6 is determined by this tangency condition (Sharipova, 2024).

The phase space is the space of oriented chords on $p_{k-1}$7. In the planar smooth setting, with $p_{k-1}$8 a positively oriented parametrization, one may write

$p_{k-1}$9

and define

$p_{k+1}$0

where $p_{k+1}$1 is uniquely chosen so that the tangent at $p_{k+1}$2 is parallel to the chord $p_{k+1}$3 (Baranzini et al., 8 Sep 2025).

2. Variational formulation, twist structure, and affine equivariance

The planar symplectic billiard map is an exact symplectic twist map generated by an area functional. In one normalization, the generating function is

$p_{k+1}$4

and the twist condition is

$p_{k+1}$5

For rational rotation number $p_{k+1}$6, the action of a periodic orbit is the negative of the area of the corresponding inscribed polygon, and Mather’s $p_{k+1}$7-function becomes the minimal average action of this area-type generating function (Baranzini et al., 8 Sep 2025).

An equivalent twist-map encoding uses coordinates $p_{k+1}$8 obtained from the generating function by

$p_{k+1}$9

so that symplectic billiards become exact symplectic twist maps on a strip $p_k$0. In analytic one-parameter families of strictly convex analytic domains, the induced family of symplectic billiard maps has twist interval $p_k$1, which is the range of admissible rotation numbers for the associated invariant curves and rational caustics (Fierobe et al., 2024).

Affine equivariance is a fundamental structural feature. In the planar theory, shifting the origin or applying any invertible affine map does not change the dynamics up to conjugacy; consequently all ellipses are dynamically equivalent to one another and to the disk in the affine-invariant sense relevant for symplectic billiards (Baranzini et al., 8 Sep 2025). The same affine equivariance underlies rigidity results for ellipses and explains why symplectic billiards differ conceptually from metric Birkhoff billiards, where different ellipses are not affinely indistinguishable as dynamical systems (Baracco et al., 2023).

3. Periodic trajectories, invariant curves, and Ivrii-type phenomena

A central problem is the size of the periodic set in phase space. For symplectic billiards, an Ivrii-type question asks whether the set of $p_k$2-periodic trajectories has measure zero, or at least empty interior, in the phase space of oriented chords. For strongly convex domains with smooth boundary in $p_k$3, the set of $p_k$4-periodic and $p_k$5-periodic symplectic billiard orbits has empty interior (Sharipova, 2024).

The $p_k$6-periodic case admits a geometric description. In the plane, a $p_k$7-periodic orbit $p_k$8 satisfies

$p_k$9

From the three characteristic lines through $\mathbb R^{2n}$0, one constructs three intersection points $\mathbb R^{2n}$1 and obtains parallelogram relations implying that $\mathbb R^{2n}$2 are midpoints of the sides of the triangle $\mathbb R^{2n}$3. This yields a one-to-one correspondence between $\mathbb R^{2n}$4-periodic symplectic billiard triangles and $\mathbb R^{2n}$5-periodic outer billiard triangles. In higher dimensions, the same construction gives an injective continuous map into outer billiard phase space, and a dimension-counting argument excludes an open set of $\mathbb R^{2n}$6-periodic symplectic points (Sharipova, 2024).

The $\mathbb R^{2n}$7-periodic case is governed by characteristic directions: $\mathbb R^{2n}$8 For a strongly convex domain, through any boundary point $\mathbb R^{2n}$9 there is at most one $$\ker(\omega|_{T_{p_k}\partial\Omega})=\mathbb R JN_{p_k}$$0-periodic symplectic billiard trajectory. The argument uses the uniqueness of the point $$\ker(\omega|_{T_{p_k}\partial\Omega})=\mathbb R JN_{p_k}$$1 whose characteristic direction is opposite to that at $$\ker(\omega|_{T_{p_k}\partial\Omega})=\mathbb R JN_{p_k}$$2, followed by uniqueness of the points $$\ker(\omega|_{T_{p_k}\partial\Omega})=\mathbb R JN_{p_k}$$3 whose characteristic directions are parallel to the line $$\ker(\omega|_{T_{p_k}\partial\Omega})=\mathbb R JN_{p_k}$$4. This excludes open families of $$\ker(\omega|_{T_{p_k}\partial\Omega})=\mathbb R JN_{p_k}$$5-periodic orbits (Sharipova, 2024).

Rational invariant curves are also rigid in analytic families. For an analytic one-parameter family of strictly convex analytic domains and a fixed rational rotation number $$\ker(\omega|_{T_{p_k}\partial\Omega})=\mathbb R JN_{p_k}$$6, the set of parameter values for which the associated symplectic billiard map has an $$\ker(\omega|_{T_{p_k}\partial\Omega})=\mathbb R JN_{p_k}$$7-periodic invariant curve is either finite or the whole parameter interval. In this sense, rational caustics are exceptionally fragile unless forced by a global integrable structure (Fierobe et al., 2024).

4. Integrability, ellipses, and rigidity

Ellipses are the basic integrable examples. Because symplectic billiards are affinely equivariant and the symplectic billiard in a circle is completely integrable, ellipses inherit complete integrability as affine images of the circle (Baracco et al., 2023). The converse is rigid: if the phase space of a planar symplectic billiard is foliated by continuous invariant closed curves that are not null-homotopic, then the table must be an ellipse (Baracco et al., 2023).

A partial-integrability version is also available in the centrally symmetric setting. If a centrally symmetric $$\ker(\omega|_{T_{p_k}\partial\Omega})=\mathbb R JN_{p_k}$$8 strongly convex domain has a simple invariant curve of rotation number $$\ker(\omega|_{T_{p_k}\partial\Omega})=\mathbb R JN_{p_k}$$9, consisting entirely of $(\mathbb R^{2n},\omega)$0-periodic orbits and winding once around the phase space, and if one of the two regions between that curve and the boundary of phase space is foliated by continuous invariant closed curves, then the boundary is an ellipse. The proof reduces the problem directly to the totally integrable case by exploiting the symmetry of the area generating function and the special geometry of $(\mathbb R^{2n},\omega)$1-periodic parallelogram orbits (Baracco et al., 2024).

Near ellipses, rational integrability is locally rigid. If a planar domain is sufficiently $(\mathbb R^{2n},\omega)$2-close and sufficiently $(\mathbb R^{2n},\omega)$3-close to an ellipse, and if for every $(\mathbb R^{2n},\omega)$4 the symplectic billiard map admits an invariant curve of rotation number $(\mathbb R^{2n},\omega)$5 consisting entirely of $(\mathbb R^{2n},\omega)$6-periodic orbits, then the domain is itself an ellipse (Tsodikovich, 15 Jan 2025). The mechanism is Fourier-analytic: rational integrability forces strong constraints on the Fourier coefficients of the deformation of the boundary, while the low modes correspond precisely to infinitesimal deformations through ellipses.

Spectral rigidity enters through the area spectrum and through Mather’s $(\mathbb R^{2n},\omega)$7-function. For axially symmetric $(\mathbb R^{2n},\omega)$8-smooth one-parameter families, an area-spectral deformation near an ellipse is trivial up to area-preserving affine transformations, and in the general axially symmetric setting any sufficiently smooth area-spectral family is tangent to a finite-dimensional space of deformations (Fierobe et al., 2024). In Aubry–Mather terms, symplectic billiards satisfy the pointwise inequality

$(\mathbb R^{2n},\omega)$9

and equality at a single rotation number forces $\omega(u,v)=\sum_{i=1}^n (u_{q_i}v_{p_i}-u_{p_i}v_{q_i}) =\langle Ju,v\rangle, \qquad J=\begin{pmatrix}0&I_n\-I_n&0\end{pmatrix}.$0 to be an ellipse. At rational $\omega(u,v)=\sum_{i=1}^n (u_{q_i}v_{p_i}-u_{p_i}v_{q_i}) =\langle Ju,v\rangle, \qquad J=\begin{pmatrix}0&I_n\-I_n&0\end{pmatrix}.$1, this reproduces Sas’s inequality for maximal areas of inscribed $\omega(u,v)=\sum_{i=1}^n (u_{q_i}v_{p_i}-u_{p_i}v_{q_i}) =\langle Ju,v\rangle, \qquad J=\begin{pmatrix}0&I_n\-I_n&0\end{pmatrix}.$2-gons (Baranzini et al., 8 Sep 2025).

5. Higher dimensions, Minkowski formulations, and polygonal models

Higher-dimensional symplectic billiards retain the characteristic-line reflection law, and the low-period Ivrii-type results extend to all dimensions through the geometric arguments already described (Sharipova, 2024). A deeper conceptual extension identifies symplectic billiards with a special case of Minkowski billiards. In the canonical symplectic vector space $\omega(u,v)=\sum_{i=1}^n (u_{q_i}v_{p_i}-u_{p_i}v_{q_i}) =\langle Ju,v\rangle, \qquad J=\begin{pmatrix}0&I_n\-I_n&0\end{pmatrix}.$3, the Minkowski billiard map admits a symplectic reduction description; after identifying $\omega(u,v)=\sum_{i=1}^n (u_{q_i}v_{p_i}-u_{p_i}v_{q_i}) =\langle Ju,v\rangle, \qquad J=\begin{pmatrix}0&I_n\-I_n&0\end{pmatrix}.$4 with $\omega(u,v)=\sum_{i=1}^n (u_{q_i}v_{p_i}-u_{p_i}v_{q_i}) =\langle Ju,v\rangle, \qquad J=\begin{pmatrix}0&I_n\-I_n&0\end{pmatrix}.$5 via the symplectic form, one obtains a symplectic version of Minkowski billiards. When the two hypersurfaces coincide, the associated Minkowski billiard map $\omega(u,v)=\sum_{i=1}^n (u_{q_i}v_{p_i}-u_{p_i}v_{q_i}) =\langle Ju,v\rangle, \qquad J=\begin{pmatrix}0&I_n\-I_n&0\end{pmatrix}.$6 satisfies

$\omega(u,v)=\sum_{i=1}^n (u_{q_i}v_{p_i}-u_{p_i}v_{q_i}) =\langle Ju,v\rangle, \qquad J=\begin{pmatrix}0&I_n\-I_n&0\end{pmatrix}.$7

so the symplectic billiard map is a “square root” of the symplectic Minkowski billiard map (Albers et al., 7 Jul 2026). This viewpoint yields new multiplicity bounds: in dimension $\omega(u,v)=\sum_{i=1}^n (u_{q_i}v_{p_i}-u_{p_i}v_{q_i}) =\langle Ju,v\rangle, \qquad J=\begin{pmatrix}0&I_n\-I_n&0\end{pmatrix}.$8, for every prime $\omega(u,v)=\sum_{i=1}^n (u_{q_i}v_{p_i}-u_{p_i}v_{q_i}) =\langle Ju,v\rangle, \qquad J=\begin{pmatrix}0&I_n\-I_n&0\end{pmatrix}.$9, there exist at least $X\subset \mathbb R^{2n}$0 geometrically distinct $X\subset \mathbb R^{2n}$1-periodic symplectic billiard orbits (Albers et al., 7 Jul 2026).

Polygonal symplectic billiards replace smooth tangent lines by side directions. For a convex polygon $X\subset \mathbb R^{2n}$2 with edge vectors $X\subset \mathbb R^{2n}$3, the reduced phase space is a union of rectangles $X\subset \mathbb R^{2n}$4 with $X\subset \mathbb R^{2n}$5, endowed with the area form

$X\subset \mathbb R^{2n}$6

and on each rectangle the map has the affine form

$X\subset \mathbb R^{2n}$7

The discontinuity set is a union of horizontal and vertical line segments, and its complement decomposes into tiles, each tile being a rectangle corresponding to a periodic symbolic orbit. Explicit examples include the Quad, in which all orbits have periods $X\subset \mathbb R^{2n}$8 or $X\subset \mathbb R^{2n}$9; the tall Penthouse, in which all orbits have periods $p_{k-1},p_k,p_{k+1}\in \partial\Omega$00, $p_{k-1},p_k,p_{k+1}\in \partial\Omega$01, or $p_{k-1},p_k,p_{k+1}\in \partial\Omega$02; and lattice hexagons with parallel opposite sides, in which all orbits are periodic with periods bounded above by $p_{k-1},p_k,p_{k+1}\in \partial\Omega$03 for an explicit combinatorial quantity $p_{k-1},p_k,p_{k+1}\in \partial\Omega$04. The paper also formulates full-periodicity conjectures for Hexhouses and for certain special octagons (Albers et al., 2019).

A two-table polygonal analogue leads to symplectic tiling billiards. For transverse tilings $p_{k-1},p_k,p_{k+1}\in \partial\Omega$05 of the plane, the dynamics acts on pairs of particles on tiling edges by parallelism rules modeled on polygonal symplectic billiards. In the special case of oriented weaves of $p_{k-1},p_k,p_{k+1}\in \partial\Omega$06-sunbursts, there is a unique phase modification making all orbits left-convex. This construction yields an elementary correspondence between convex equilateral and convex equiangular polygons and, through Thurston’s “Shapes of Polyhedra,” produces hyperbolic structures on linkage moduli spaces; in particular, the configuration space of the hexagonal planar linkage with unit-length rods is described as a $p_{k-1},p_k,p_{k+1}\in \partial\Omega$07-cusped hyperbolic $p_{k-1},p_k,p_{k+1}\in \partial\Omega$08-manifold tiled by $p_{k-1},p_k,p_{k+1}\in \partial\Omega$09 regular ideal octahedra (Schwartz, 2023).

Related Hamiltonian product-body billiards on $p_{k-1},p_k,p_{k+1}\in \partial\Omega$10 are defined by the characteristic line field of the standard symplectic form on the boundary of $p_{k-1},p_k,p_{k+1}\in \partial\Omega$11. Their projections to $p_{k-1},p_k,p_{k+1}\in \partial\Omega$12 are $p_{k-1},p_k,p_{k+1}\in \partial\Omega$13-billiards, which coincide with Minkowski Finsler billiards when $p_{k-1},p_k,p_{k+1}\in \partial\Omega$14 is centrally symmetric. In that broader setting, the reflection law is projective if and only if $p_{k-1},p_k,p_{k+1}\in \partial\Omega$15 is an ellipsoid; equivalently, all such $p_{k-1},p_k,p_{k+1}\in \partial\Omega$16-billiards are simultaneously affine equivalent to Euclidean billiards precisely in the ellipsoidal case (Glutsyuk, 2024).

6. Dissipative and symplectic-topological directions

A dissipative variant of symplectic billiards is obtained by composing the conservative twist map $p_{k-1},p_k,p_{k+1}\in \partial\Omega$17 with the fiber contraction

$p_{k-1},p_k,p_{k+1}\in \partial\Omega$18

thereby producing a conformally symplectic map $p_{k-1},p_k,p_{k+1}\in \partial\Omega$19 with area-contraction factor $p_{k-1},p_k,p_{k+1}\in \partial\Omega$20. Its global invariant separator is the Birkhoff attractor. For strong dissipation, the Birkhoff attractor coincides with the global attractor and is a normally contracted $p_{k-1},p_k,p_{k+1}\in \partial\Omega$21 graph over the zero section; for sufficiently stronger dissipation it becomes a $p_{k-1},p_k,p_{k+1}\in \partial\Omega$22 graph converging in $p_{k-1},p_k,p_{k+1}\in \partial\Omega$23 to the zero section as $p_{k-1},p_k,p_{k+1}\in \partial\Omega$24 (Baracco et al., 16 Sep 2025). For mild dissipation in a centrally symmetric domain with an instability region around the zero section, the attractor becomes an indecomposable continuum and the restricted dynamics has positive topological entropy (Baracco et al., 16 Sep 2025).

A broader symplectic-topological program studies billiard maps through Hamiltonian metrics and invariants. For smooth strictly convex billiard tables in the plane, the billiard ball map lies in a Hofer-geometric coset of $p_{k-1},p_k,p_{k+1}\in \partial\Omega$25, and the Hofer distance between two billiard maps satisfies

$p_{k-1},p_k,p_{k+1}\in \partial\Omega$26

where $p_{k-1},p_k,p_{k+1}\in \partial\Omega$27 is a geometric path metric on billiard tables. Although this concerns classical billiard ball maps rather than the symplectic billiard model itself, it exemplifies the increasing use of Hofer geometry, displacement energy, and Floer-theoretic ideas in billiard dynamics (Berezovik et al., 7 Jul 2025).

Several open problems remain central. For symplectic billiards beyond periods $p_{k-1},p_k,p_{k+1}\in \partial\Omega$28 and $p_{k-1},p_k,p_{k+1}\in \partial\Omega$29, it is natural to ask whether the set of $p_{k-1},p_k,p_{k+1}\in \partial\Omega$30-periodic orbits has empty interior for all $p_{k-1},p_k,p_{k+1}\in \partial\Omega$31, and whether a full Ivrii-type measure-zero statement holds for all periods (Sharipova, 2024). The analytic-family theory suggests that rational caustics of a fixed rotation number are generically absent unless enforced across an entire family, but a geometric characterization of domains carrying extensive families of rational invariant curves is still incomplete (Fierobe et al., 2024). These questions connect periodic orbit geometry, convexity, integrability, and symplectic topology in a way that has become characteristic of the modern theory of symplectic billiards.