Hamiltonian monodromy and the spherical pendulum

Intuition Beginner

The Arnold-Liouville theorem 05.02.04 promises that near any one of its doughnut-shaped invariant surfaces, an integrable system has a tidy set of coordinates: a few "action" dials that label which doughnut you are on, and matching "angle" dials that say where you are around it. The promise is local. It holds on a patch around each doughnut.

Hamiltonian monodromy is the discovery that you cannot always glue those patches into one chart that covers the whole family of doughnuts at once. Sometimes the family is twisted.

The simplest twist comes from a single bad point in the middle of the energy values, where the doughnut pinches. Go around that point.

Visual Beginner

Picture the set of allowed energy-and-spin values as a flat sheet. Most points on the sheet sit above a doughnut. One special point is a puncture: the doughnut there is pinched.

Now carry a labelled pair of loops — one going the short way around a doughnut, one the long way — slowly along a circle that surrounds the puncture. When you return to the start, the short loop is unchanged, but the long loop has picked up an extra copy of the short one. The labels no longer match where they began.

That shearing of the labels is the monodromy. It is the picture-proof that no single global labelling can exist.

Worked example Beginner

The spherical pendulum is a bead sliding on a frictionless sphere under gravity. It has two conserved quantities: the energy, and the spin around the vertical axis through the support point.

Plot every value of (energy, spin) the bead can have. The plot is a region in the plane, and almost every point inside it corresponds to one doughnut of motions. One point stands out: the value where the bead would balance, motionless, straight up at the top. That balanced state is unstable, and it pinches its doughnut to a point.

Drag the loop-labels once around that one special value. The long loop comes back carrying one extra short loop. Written as a bookkeeping rule, the new labels are the old ones with a single unit of shear added.

What this tells us: the spherical pendulum, perfectly solvable as it is, has no one global set of action-angle dials. The unstable top is to blame.

Check your understanding Beginner

Formal definition Intermediate+

Let $(M^4,\omega )$ be a symplectic manifold carrying a Liouville-integrable system of two degrees of freedom 05.02.03, with two independent Poisson-commuting integrals assembled into the energy-momentum map $F=(H,J):M\to {\mathbb{R}}^2$. Let $R\subset {\mathbb{R}}^2$ be the set of regular values of $F$ — those $c$ at which $dH$ and $dJ$ are independent on all of $F^{-1}(c)$ — and assume each regular fibre is compact and connected. The Arnold-Liouville theorem 05.02.04 makes every regular fibre a Lagrangian $2$-torus $T_c=F^{-1}(c)$, so $F:F^{-1}(R)\to R$ is a bundle of $2$-tori.

Over $R$ sits the period lattice, the bundle of first integral homologies $$ \mathcal{L} ;=; \bigsqcup_{c \in R} H_1(T_c; \mathbb{Z}) ;\cong; \bigsqcup_{c\in R}\mathbb{Z}^2 , $$ a local system of free abelian groups of rank $2$. Choosing a basis $({\gamma }_1(c),{\gamma }_2(c))$ at one regular value and parallel-transporting it along a loop $\sigma :[0,1]\to R$ returns a basis differing from the original by an automorphism of ${\mathbb{Z}}^2$.

The monodromy representation is the resulting homomorphism $$ \mu : \pi_1(R, c_0) \longrightarrow \operatorname{Aut}\big(H_1(T_{c_0};\mathbb{Z})\big) \cong GL(2, \mathbb{Z}), $$ $$ \mu([\sigma]) = \text{(parallel transport of the basis around } \sigma\text{)}. $$ Because action coordinates pair the lattice with the symplectic form, $\mu$ in fact lands in $SL(2,\mathbb{Z})$. When some $\mu ([\sigma ])\ne \mathrm{id}$, the bundle $\mathcal{L}$ admits no global basis, and consequently no globally single-valued action variables exist ^{[Duistermaat 1980]}.

Key theorem with proof Intermediate+

Theorem (Duistermaat 1980). Let $F=(H,J):M^4\to {\mathbb{R}}^2$ be a two-degree-of-freedom integrable system with compact connected regular fibres, and let $R$ be the set of regular values. Suppose $R = \mathbb{R}^2 \setminus {c_}$forasingleisolatedcriticalvalue$c_*$whosefibre$F^{-1}(c_*)$carriesone\ast \ast focus-focus\ast \ast criticalpoint.Let$\sigma$beasmallloopin$R$encircling$c_*$once,positivelyoriented.Theninasuitablebasis$(\gamma_1, \gamma_2)$of$H_1(T;\mathbb{Z})$ the monodromy is* $$ \mu([\sigma]) ;=; \begin{pmatrix} 1 & 1 \ 0 & 1 \end{pmatrix} \in SL(2,\mathbb{Z}), $$ and therefore the torus bundle over $R$ admits no global action coordinates.

Proof.

Step 1: the focus-focus normal form. By the Eliasson linearisation theorem, near the focus-focus point the system is symplectically equivalent to the quadratic model on $({\mathbb{R}}^4,d{x}_1\wedge d{\xi }_1+d{x}_2\wedge d{\xi }_2)$ with commuting integrals $$ q_1 = x_1\xi_1 + x_2\xi_2, \qquad q_2 = x_1\xi_2 - x_2\xi_1 . $$ The pair $(q_1,q_2)$ has an isolated singular value at the origin, and the local fibre over the origin is a pinched torus — a $2$-torus with one cycle collapsed to the singular point.

Step 2: a distinguished cycle. On the model put $w=x_1+i{x}_2$ and $z={\xi }_1+i{\xi }_2$, so $q_1+i{q}_2=\bar{z}w$. A regular fibre near the origin is $\{\bar{z}w=c\}$ for small $c\ne 0$. Holding $|w|$ fixed and rotating the phase of $w$ through $2\pi$ traces a loop ${\gamma }_1$ on the fibre; this cycle survives the limit $c\to 0$ shrinking to the singular point, so it is the vanishing cycle. A second cycle ${\gamma }_2$ — any loop closing off the complementary direction — completes a basis with ${\gamma }_1$.

Step 3: holonomy of the second cycle. Transport the basis along $\sigma$, a loop of radius $\rho$ about $c_{\ast }=0$ in the value plane. The vanishing cycle ${\gamma }_1$ is monodromy-invariant: it is intrinsically attached to the pinch and returns to itself, so $\mu ({\gamma }_1)={\gamma }_1$. For ${\gamma }_2$, compute the variation by the Picard-Lefschetz formula governing an isolated quadratic (focus-focus) degeneration: each circuit adds one copy of the vanishing cycle, $$ \mu(\gamma_2) = \gamma_2 + \langle \gamma_2, \gamma_1\rangle, \gamma_1 = \gamma_2 + \gamma_1, $$ where the intersection number is normalised to $⟨{\gamma }_2,{\gamma }_1⟩=1$ by orientation.

Step 4: read off the matrix. In the ordered basis $({\gamma }_1,{\gamma }_2)$ the two relations $\mu ({\gamma }_1)={\gamma }_1$ and $\mu ({\gamma }_2)={\gamma }_1+{\gamma }_2$ give exactly $$ \mu([\sigma]) = \begin{pmatrix} 1 & 1 \ 0 & 1\end{pmatrix}. $$ This matrix has infinite order, so no power is the identity; the local system $\mathcal{L}$ has no ${\pi }_1(R)$-invariant basis. A global action map would supply one, so none exists. $\square$

Bridge. This builds toward the global integrable-systems theory and appears again in the semiclassical units, because the shear matrix here is exactly the obstruction whose existence the synthesis of 05.02.04 only names. The foundational reason action-angle coordinates are local is now visible: the central insight is that the period lattice is a twisted local system, and putting these together, the local normal form of 05.02.04 and the global holonomy computed here are dual sides of one structure — the bridge is that a vanishing cycle attached to a focus-focus pinch generalises the smooth torus bundle into a twisted one, and this is exactly the mechanics counterpart of the Picard-Lefschetz monodromy of an isolated singularity.

Exercises Intermediate+

Advanced results Master

The monodromy theorem is the local model for a wider phenomenon. Any isolated singular fibre of focus-focus type, with $k$ focus-focus points stacked on one fibre, produces monodromy $\left(\begin{array}{cc}1& k\\ 0& 1\end{array}\right)$ around it: the contributions of the individual pinches add, because each contributes one Picard-Lefschetz shear along the common vanishing cycle. The number $k$ is a symplectic invariant of the singular fibre, computable from the Eliasson normal form, and it is the integer that distinguishes, for instance, the spherical pendulum ($k=1$) from systems with coincident foci.

The same matrix governs the quantum system. Bohr-Sommerfeld quantisation places joint eigenvalues of the commuting operators $(\hat{H},\hat{J})$ at a lattice of action values; the classical monodromy reappears as the global obstruction to labelling this joint spectrum by a single pair of integer quantum numbers. Vũ Ngọc made this precise: the asymptotic lattice of eigenvalues has a dislocation whose Burgers-like defect is exactly the classical monodromy matrix, so the integer $k$ is measurable in a spectrum ^{[Vũ Ngọc 1999]}.

Beyond focus-focus lie further refinements. The Duistermaat-Heckman measure controls how the symplectic volume of reduced spaces varies, and its non-smoothness across singular values is governed by the same fibre data. Fractional and "scattering" monodromy extend the integer holonomy to rational matrices and to non-compact value sets, capturing the hydrogen atom in crossed electric and magnetic fields and the swing-spring (Fermi) resonance. In every case the organising principle is unchanged: a singular fibre punches a hole in the regular-value set, and the period lattice acquires holonomy around it.

Synthesis. Hamiltonian monodromy is the foundational reason the Arnold-Liouville theorem 05.02.04 is stated locally, and it appears again across the strand as a single recurring mechanism. The central insight is that the period lattice $H_1(T_c;\mathbb{Z})$ is a local system over the regular values, and putting these together with the focus-focus normal form shows that the integer monodromy is exactly the Picard-Lefschetz shear of an isolated singularity — this is exactly the mechanics image of vanishing-cycle theory, and it generalises the smooth torus bundle of an integrable system 05.02.03 into a twisted one. The classical shear is dual to the quantum dislocation in the joint spectrum, so the same invariant $k$ builds toward both the global obstruction to action variables and the obstruction to global quantum numbers; the bridge is that local normal form and global holonomy are two readings of one geometric object, the singular Lagrangian fibration.

Full proof set Master

Proposition. The focus-focus monodromy matrix $\left(\begin{array}{cc}1& 1\\ 0& 1\end{array}\right)$ has infinite order in $SL(2,\mathbb{Z})$, and consequently the period lattice over $R = \mathbb{R}^2\setminus{c_}$ admits no global basis.*

Proof. By induction, ${\left(\begin{array}{cc}1& 1\\ 0& 1\end{array}\right)}^n=\left(\begin{array}{cc}1& n\\ 0& 1\end{array}\right)$: the base case $n=1$ is immediate, and $$ \begin{pmatrix}1&n\0&1\end{pmatrix}\begin{pmatrix}1&1\0&1\end{pmatrix} = \begin{pmatrix}1&n+1\0&1\end{pmatrix}. $$ The off-diagonal entry $n$ is never zero for $n\ne 0$, so no positive power equals the identity, and the element has infinite order. Now ${\pi }_1(R)\cong \mathbb{Z}$, generated by the loop $\sigma$ about $c_{\ast }$, and the monodromy representation sends the generator to this infinite-order matrix, hence is injective with image the infinite cyclic group \{\begin{psmallmatrix}1&n\\0&1\end{psmallmatrix}\}. A global basis of the lattice would be a section of the frame bundle fixed by all of ${\pi }_1(R)$, that is a common eigenframe of every \begin{psmallmatrix}1&n\\0&1\end{psmallmatrix} with integer eigenvalue $1$ in each coordinate. The only common fixed vector is along the vanishing cycle ${\gamma }_1$; the complementary generator is shifted by $n{\gamma }_1$ and cannot be fixed. A free basis needs two independent invariant vectors, so none exists. $\square$

Proposition. The vanishing cycle is the unique (up to sign) primitive monodromy-invariant class in $H_1(T;\mathbb{Z})$.

Proof. A class $a{\gamma }_1+b{\gamma }_2$ is fixed by \begin{psmallmatrix}1&1\\0&1\end{psmallmatrix} iff $a{\gamma }_1+b({\gamma }_1+{\gamma }_2)=a{\gamma }_1+b{\gamma }_2$, which forces $b{\gamma }_1=0$, hence $b=0$. The fixed sublattice is therefore $\mathbb{Z}{\gamma }_1$, and its primitive generators are $\pm {\gamma }_1$. $\square$

Connections Master

• The whole construction is the global completion of 05.02.04: the Arnold-Liouville theorem builds action-angle coordinates on a patch around each Liouville torus, and Hamiltonian monodromy is the precise obstruction to assembling those patches into one global chart — the central insight that 05.02.04's synthesis only names is computed here.

• The underlying object is the integrable system of 05.02.03, whose regular fibres form the torus bundle; the energy-momentum map and its period lattice are read off the cotangent-bundle Liouville form of 05.02.05, which supplies the primitive whose period integrals are the action variables that fail to globalise.

• The mechanism is dual to the semiclassical units on Bohr-Sommerfeld quantisation 12.07.04, where the classical monodromy matrix reappears as a dislocation in the joint spectrum, so the same invariant obstructs both global action variables and global quantum numbers, linking the symplectic and quantum strands through one integer.

Historical & philosophical context Master

The local theory of action-angle variables was completed by Nekhoroshev in 1972, who gave the modern statement of the Arnold-Liouville construction on a neighbourhood of a single torus ^{[Nekhoroshev 1972]}. The decisive global step came eight years later, when Duistermaat asked whether those local coordinates always patch together and found that they need not: his 1980 paper isolated the topological monodromy of the torus bundle as the obstruction and exhibited the spherical pendulum — a textbook system known since the eighteenth century — as the elementary counterexample ^{[Duistermaat 1980]}. The conceptual surprise was that completely solvable, classical, and even pedagogical systems hide a global topological defect invisible to any local computation. The story closed a loop with singularity theory: the focus-focus monodromy is the mechanical face of the Picard-Lefschetz formula for an isolated complex singularity, and through the SYZ picture it reappears on the mirror-symmetry side as the monodromy of special Lagrangian fibrations. Cushman and Bates later turned the phenomenon into a systematic chapter of classical mechanics, and Vũ Ngọc supplied its quantum shadow, making monodromy a measurable feature of molecular spectra rather than a geometric curiosity.

Bibliography Master

@article{Duistermaat1980GlobalActionAngle,
  author  = {Duistermaat, J. J.},
  title   = {On global action-angle coordinates},
  journal = {Communications on Pure and Applied Mathematics},
  volume  = {33},
  number  = {6},
  year    = {1980},
  pages   = {687--706}
}

@book{CushmanBates1997Global,
  author    = {Cushman, Richard H. and Bates, Larry M.},
  title     = {Global Aspects of Classical Integrable Systems},
  publisher = {Birkh\"auser},
  year      = {1997}
}

@article{Nekhoroshev1972ActionAngle,
  author  = {Nekhoroshev, N. N.},
  title   = {Action-angle variables and their generalizations},
  journal = {Transactions of the Moscow Mathematical Society},
  volume  = {26},
  year    = {1972},
  pages   = {180--198}
}

@article{VuNgoc1999QuantumMonodromy,
  author  = {V\~u Ng\d{o}c, San},
  title   = {Quantum monodromy in integrable systems},
  journal = {Communications in Mathematical Physics},
  volume  = {203},
  number  = {2},
  year    = {1999},
  pages   = {465--479}
}

@book{ArnoldMechanics,
  author    = {Arnold, V. I.},
  title     = {Mathematical Methods of Classical Mechanics},
  publisher = {Springer},
  year      = {1989}
}