Focus–Focus Singularities in Hamiltonian Systems

• Focus–focus singularities are defined as nondegenerate critical points in integrable Hamiltonian systems, exhibiting pinched torus fibrations and characteristic monodromy.
• They obstruct the existence of global action–angle coordinates and induce singular integral affine structures, with local normal forms described by the Eliasson–Vey theorem.
• Their classification utilizes invariants such as Taylor series expansions, discrete moduli, and twisting indices to capture both local dynamics and global topology.

A focus–focus singularity is a distinguished, non-degenerate singular point of an integrable Hamiltonian system—in simplest form, a two-degree-of-freedom symplectic manifold $(M^4,\omega)$ equipped with a Poisson-commuting pair $F=(H_1,H_2): M \to \mathbb{R}^2$. Such singularities exhibit complex dynamical and topological features, notably pinched torus fibrations, the emergence of singular integral affine structures with nontrivial monodromy, and a hierarchy of smooth, topological, and symplectic invariants. Their presence fundamentally obstructs the existence of global action–angle coordinates and underpins key phenomena in semitoric systems, singular affine geometry, and even applications in singular optics.

1. Local Normal Form and Geometry of Focus–Focus Singularities

Let $p\in M$ be an isolated nondegenerate “focus–focus” singular point (Williamson type $(0,0,1)$). By the Eliasson–Vey theorem, there exist local symplectic coordinates $(x_1,y_1,x_2,y_2)$ with $\omega=dx_1\wedge dy_1+dx_2\wedge dy_2$, in which the two first integrals depend only on the quadratic invariants

$$q_1 = x_1y_2 - x_2y_1,\qquad q_2 = x_1y_1 + x_2y_2.$$

A canonical form near $p$ is $F=(q_1, q_2)$; or $z=uv$ in complex variables $F=(H_1,H_2): M \to \mathbb{R}^2$0, $F=(H_1,H_2): M \to \mathbb{R}^2$1 (Ngoc et al., 2011, Bolsinov et al., 2017). The map $F=(H_1,H_2): M \to \mathbb{R}^2$2 is, up to smooth change, $F=(H_1,H_2): M \to \mathbb{R}^2$3 near $F=(H_1,H_2): M \to \mathbb{R}^2$4.

The singular fiber over $F=(H_1,H_2): M \to \mathbb{R}^2$5 is a pinched torus: a two-torus where one (or more, in higher “complexity” $F=(H_1,H_2): M \to \mathbb{R}^2$6) cycles is collapsed to a point. For $F=(H_1,H_2): M \to \mathbb{R}^2$7 (simple focus–focus), the local topology is a single nodal torus. For $F=(H_1,H_2): M \to \mathbb{R}^2$8 (multi-pinched), the singular fiber is a chain of Lagrangian spheres intersecting transversely. The regular nearby fibers are tori, but the fibration exhibits a vanishing cycle collapsing at the focus–focus point (Meulenaere et al., 2019, Smirnov, 2013, Pelayo et al., 2018).

2. Singular Affine Structures and Monodromy

On the base $F=(H_1,H_2): M \to \mathbb{R}^2$9, the image admits a singular integral affine structure: away from the critical value the base inherits action–angle coordinates $p\in M$0; at the singular value, the affine structure has monodromy. Explicitly, action functions near a regular value $p\in M$1 can be defined via integrals of a Liouville 1-form along a basis of $p\in M$2, but near the focus–focus critical value $p\in M$3, one action, $p\in M$4, is multi-valued, and branches differ by the monodromy jump: $p\in M$5 for $p\in M$6, with $p\in M$7 the number of pinches (Ratiu et al., 2017, Tang, 2024).

Analytic continuation of the period lattice around the singular value transforms via the monodromy matrix

$p\in M$8

in $p\in M$9 (Ratiu et al., 2017, Pelayo et al., 2018). For $(0,0,1)$0, this is the classical focus–focus monodromy. The base can be globally constructed by cutting out a sector $(0,0,1)$1 and gluing sides by $(0,0,1)$2; this realization, known as the developing map, encodes the singular affine structure (Ratiu et al., 2017, Tang, 2024).

3. Convexity and Global Topological Phenomena

Locally, every sufficiently small “focus box”

$(0,0,1)$3

is convex in the singular affine sense: any two points in $(0,0,1)$4 are connected by an affine geodesic in $(0,0,1)$5 (the “focus–box convexity” theorem). Globally, the total monodromy of the system controls convexity: with one focus–focus point (or simple arrangements), the base $(0,0,1)$6 remains globally convex under reasonable topological conditions; but with multiple interacting focus–focus points, global convexity can fail even on manifolds homeomorphic to $(0,0,1)$7 (Ratiu et al., 2017). The most striking example is the “integral–affine black hole,” constructed by gluing an eight-petal region with eight focus–focus singularities so that no affine geodesic from the center escapes the region, illustrating drastic global effects of large monodromy.

4. Classification, Invariants, and Multi-Pinched Cases

Local and Semi-Global Classification

A focus–focus singularity is completely determined up to fiberwise symplectomorphism by—

• its formal Taylor expansion invariants (“Vũ Ngọc series”) encoding the regularized action, i.e., the terms in the expansion of the action function beyond the logarithmic (singular) term (Pelayo et al., 2018, Ngoc et al., 2011, Meulenaere et al., 2019),
• discrete data: number of pinches ($(0,0,1)$8), orientation, and flatness conditions,
• and, in the case of multi-pinched (complexity $(0,0,1)$9), by $(x_1,y_1,x_2,y_2)$0 independent invariants modulo $(x_1,y_1,x_2,y_2)$1 symmetry, as per the solution to Vũ Ngọc's conjecture (Pelayo et al., 2018).

For a 2-pinched singularity (double focus–focus), the moduli space of smooth structures is one-dimensional, parameterized by a modulus $(x_1,y_1,x_2,y_2)$2, arising from the gluing of local models by smooth transition jets (Bolsinov et al., 2017). These smooth invariants are essential for distinguishing homeomorphic but non-diffeomorphic focus–focus singularities.

Global Symplectic Invariants

In the context of semitoric systems, a system with $(x_1,y_1,x_2,y_2)$3 focus–focus points is classified by five invariants per point:

• the number $(x_1,y_1,x_2,y_2)$4 of focus–focus singularities,
• the semitoric polygon (with vertical cuts at focus values),
• the Taylor series invariant (symplectic),
• the height/volume invariant (vertical location of singularities in the polygon),
• and the twisting index (how local toric models are glued globally).

For simple focus–focus fibers ($(x_1,y_1,x_2,y_2)$5), the affine structure suffices for classification. For $(x_1,y_1,x_2,y_2)$6, the collection of $(x_1,y_1,x_2,y_2)$7 Taylor series and gluing data (transition maps) is required; the same affine structure may arise from non-isomorphic systems with different gluing data (Tang, 2024, Pelayo et al., 2018, Bolsinov et al., 2017).

5. Dynamical and Affine Features, and Applications

Action–Angle Failure and Nontrivial Dynamics

The breakdown of the Liouville–Arnold theorem at focus–focus points manifests in the monodromic behavior of action integrals. In a neighborhood of a focus–focus point, the multi-valuedness is given by a complex logarithm: $(x_1,y_1,x_2,y_2)$8 with $(x_1,y_1,x_2,y_2)$9, and $\omega=dx_1\wedge dy_1+dx_2\wedge dy_2$0 smooth. This creates the characteristic monodromy and logarithmic singularity in the affine base (Wacheux, 2014). The result obstructs the existence of any global action–angle system.

Convexity Failure and Black Hole Phenomena

With large or interacting monodromy, the singular affine base can support regions from which no affine geodesic can reach the boundary—integral-affine “black holes.” Local convexity (proximal to a focus–focus point) is always maintained, but it does not extend globally when monodromy subgroups are large or noncommutative (Ratiu et al., 2017).

Extensions and Physical Manifestations

Focus–focus structures have analogs beyond symplectic geometry. In optics, square-integrable paraxial fields can spontaneously form amplitude singularities in focal planes (“focus–focus singularity of the field”), with the nature of the singularity controlled by the power-law decay of the input (Aiello, 2015).

In piecewise-smooth systems (Filippov systems in $\omega=dx_1\wedge dy_1+dx_2\wedge dy_2$1), “sewed foci” represent focus–focus-like singularities with strikingly rich dynamics, including infinite-time approaches and the possibility of uncountably many distinct local phase portraits when analyticity is lost (Glendinning et al., 2023).

6. Multi-Focus–Focus Points and Symplectic Groupoid Structures

Focus–focus singularities support a canonical addition law on the regular fibers. In a Lagrangian torus fibration, this fiberwise group structure extends across the focus–focus point only as an immersed Lagrangian correspondence, not as an embedding. The self-intersection locus coincides with the triple intersection at the singular point, reflecting the topological nontriviality imposed by monodromy and the singular affine structure (Abouzaid et al., 2024).

The semi-global symplectic neighborhood around a focus–focus fiber is fully classified by the Taylor expansion of the regularized action function $\omega=dx_1\wedge dy_1+dx_2\wedge dy_2$2 (the so-called Vũ Ngọc invariant), which encodes both monodromy and higher-order dynamical data (Abouzaid et al., 2024).

7. Tabulation of Core Invariants and Classification Features

Invariant Type	Simple Focus–Focus ($\omega=dx_1\wedge dy_1+dx_2\wedge dy_2$3)	Multi-Pinched ($\omega=dx_1\wedge dy_1+dx_2\wedge dy_2$4)
Monodromy	$\omega=dx_1\wedge dy_1+dx_2\wedge dy_2$5	$\omega=dx_1\wedge dy_1+dx_2\wedge dy_2$6
Taylor Series Invariant	One formal series (Vũ Ngọc) per point	$\omega=dx_1\wedge dy_1+dx_2\wedge dy_2$7 series $\omega=dx_1\wedge dy_1+dx_2\wedge dy_2$8 (plus transition data)
Smooth Moduli	Discrete data (signs, flat diffeos)	Moduli space of dimension $\omega=dx_1\wedge dy_1+dx_2\wedge dy_2$9 (first-order) (Bolsinov et al., 2017)
Affine Polygon (Semitoric)	Marked polygon with one cut/height	Polygon, $$q_1 = x_1y_2 - x_2y_1,\qquad q_2 = x_1y_1 + x_2y_2.$$0 cuts, $$q_1 = x_1y_2 - x_2y_1,\qquad q_2 = x_1y_1 + x_2y_2.$$1 heights, $$q_1 = x_1y_2 - x_2y_1,\qquad q_2 = x_1y_1 + x_2y_2.$$2 twisting indices

• (Ratiu et al., 2017) Systematic study of convexity and affine structures in toric-focus systems; theory of integral-affine black holes.
• (Pelayo et al., 2018) Complete symplectic classification of $$q_1 = x_1y_2 - x_2y_1,\qquad q_2 = x_1y_1 + x_2y_2.$$3-pinched focus–focus fibers up to group action; solution of Vũ Ngọc’s conjecture.
• (Ngoc et al., 2011, Wacheux, 2014, Meulenaere et al., 2019, Bolsinov et al., 2017, Tang, 2024) Theories of normal form, smooth invariants, and classification in semitoric and general settings.
• (Abouzaid et al., 2024) Addition law and Lagrangian correspondence; immersion properties.
• (Aiello, 2015, Glendinning et al., 2023) Focus–focus structure in optics and non-smooth dynamics.
• (Smirnov, 2013, Hohloch et al., 2017, Alonso et al., 2020) Applications and explicit models for focus–focus fibers and semitoric invariants.

Focus–focus singularities thus form the archetypal node in singular Lagrangian torus fibrations, fundamentally shaping the geometry, dynamics, and classification of integrable Hamiltonian systems and their associated integral affine structures.