Action Degeneracy: Dynamics, KAM, and Applications

• Action degeneracy is the failure of standard nondegeneracy conditions, leading to obstructions in continuation, uniqueness, and controllability across various systems.
• It manifests in diverse settings—from variable-rank symplectic structures in dynamics and degenerate action–angle–angle maps in KAM theory to masked exploration in principal-agent games.
• The phenomenon also appears via symmetry-induced multiplicities in rational Painlevé V solutions and discrete degeneracies in holographic black-hole saddles, yielding distinct physical and computational outcomes.

Action degeneracy is not a single invariant notion across contemporary research. In the cited literature, the expression designates distinct failures of standard nondegeneracy assumptions: variable rank of a presymplectic structure in first-order dynamics, a rank-deficient frequency map in three-dimensional action–angle–angle maps, non-implementability of actions in contextual principal-agent games, symmetry-induced multiplicity of rational Painlevé V solutions, and discrete one-form-symmetry-related multiplicity of holographic black-hole saddles (Micheli et al., 2012, Vaidya et al., 2014, Feng et al., 21 Oct 2025, Aratyn et al., 2023, Cassani et al., 1 Jun 2026). What unifies these usages is not a common formal definition, but the role of degeneracy as an obstruction to generic continuation, uniqueness, or controllability.

1. Terminological scope

In the literature considered here, “action degeneracy” is domain-specific. The same phrase can refer to a structural property of the equations of motion, of a frequency map, of an optimization/learning interface, of a symmetry orbit, or of a saddle-point expansion.

Domain	Degeneracy notion	Consequence
First-order dynamical systems	$\operatorname{rank} F_{ij}$ not constant	phase-space sectorization
3D volume-preserving maps	one frozen angle, one action, two angles	modified KAM persistence
Principal-agent learning	action not implementable on $\Theta_t$	masked exploration
Rational Painlevé V	distinct symmetry-generated solutions share parameters	two-fold degeneracy
Gravitational index	symmetry-related saddles share on-shell action	$\log |G|$ correction

A common misconception is to treat these meanings as interchangeable. They are not. In the KAM setting, degeneracy is a twist/rank deficiency of the frequency map rather than a loss of rank of a symplectic two-form. In the Painlevé V setting, the paper explicitly characterizes the effect as a group-action phenomenon rather than a spectral one. In the holographic index, degeneracy is not a continuum of zero-modes, but a finite set of symmetry-related bulk sectors.

2. Variable-rank symplectic structure and intrinsic degeneracy

For finite-dimensional systems in first-order form,

$S[z]=\int (A_i(z)\dot z^i + A_0(z))\,dt,$

the presymplectic two-form is

$F=dA,\qquad F_{ij}:=\partial_i A_j-\partial_j A_i.$

A system is degenerate when the rank of $F_{ij}$ is not constant throughout phase space, equivalently when

$$\Sigma := \{z\in \Gamma \mid \Delta(z)=0\},\qquad \Delta(z):=\det[F_{ij}(z)].$$

Away from $\Sigma$, $F_{ij}$ is invertible and the Hamiltonian flow is

$\dot z^i = F^{ij}(z)\partial_j H(z).$

The zero set $\Theta_t$0 is invariant under nonsingular coordinate changes because $\Theta_t$1 transforms as a pseudoscalar, so degeneracy cannot be removed by smooth redefinitions of variables (Micheli et al., 2012).

The paper distinguishes reducible from irreducible degeneracy. In a two-dimensional block with

$\Theta_t$2

reducibility to a nondegenerate action is equivalent to $\Theta_t$3 being a constant of motion:

$\Theta_t$4

If this condition fails, the degeneracy is intrinsic. In that case, classical trajectories intersect the degeneracy surface, but they do not cross it smoothly. For simple zeros of $\Theta_t$5, the velocity scales as $\Theta_t$6, changes sign across $\Theta_t$7, and phase space is partitioned into dynamically disconnected regions.

Quantization preserves this sectorization for irreducible degeneracy. Since $\Theta_t$8 diverges on $\Theta_t$9, the operator algebra is well-defined only off the degeneracy surface. Self-adjointness then forces the Hilbert space to split as

$\log |G|$0

with each $\log |G|$1 supported on one nondegenerate region. The central conclusion is that there is no quantum tunnelling across degeneracy surfaces. In this sense, $\log |G|$2 is not merely a singular subset of coordinates, but a boundary separating distinct physical systems.

3. Degenerate action–angle–angle maps and KAM persistence

In three-dimensional volume-preserving dynamics with action–angle–angle coordinates, the integrable limit may contain one frozen angle. A canonical flow example is

$\log |G|$3

so one angle has zero unperturbed frequency. The map studied in the KAM result has one action $\log |G|$4 and two angles $\log |G|$5, with unperturbed form

$\log |G|$6

and perturbed form

$\log |G|$7

where the perturbations are real-analytic and $\log |G|$8-periodic in $\log |G|$9 (Vaidya et al., 2014).

Here degeneracy has two coupled meanings. First, there is a mismatch between the number of actions and angles. Second, one angle has zero unperturbed frequency, so the frequency map has deficient rank. The map is assumed to satisfy the intersection property, namely that any torus of the form $S[z]=\int (A_i(z)\dot z^i + A_0(z))\,dt,$0 intersects its image. This plays the structural role normally associated with symplectic twist. A further assumption is the “second twist condition”

$S[z]=\int (A_i(z)\dot z^i + A_0(z))\,dt,$1

which provides convexity in the $S[z]=\int (A_i(z)\dot z^i + A_0(z))\,dt,$2 drift of the degenerate angle.

Under these assumptions, the main theorem proves persistence of a positive-measure Cantor family of two-dimensional invariant tori. They can be parameterized as

$S[z]=\int (A_i(z)\dot z^i + A_0(z))\,dt,$3

with induced dynamics

$S[z]=\int (A_i(z)\dot z^i + A_0(z))\,dt,$4

where $S[z]=\int (A_i(z)\dot z^i + A_0(z))\,dt,$5. Thus the nondegenerate angle carries quasi-periodic motion with frequency $S[z]=\int (A_i(z)\dot z^i + A_0(z))\,dt,$6, while the degenerate angle acquires an $S[z]=\int (A_i(z)\dot z^i + A_0(z))\,dt,$7 drift plus analytic correction.

The proof is non-standard because naive small-divisor estimates produce terms of order $S[z]=\int (A_i(z)\dot z^i + A_0(z))\,dt,$8. The paper therefore inserts an intermediate finite block of coordinate transformations between an initial averaging step and an infinite Moser-style KAM iteration. The purpose of this intermediate block is to reduce the action perturbation to $S[z]=\int (A_i(z)\dot z^i + A_0(z))\,dt,$9 before the infinite scheme begins. This compensates for the extra small-divisor amplification caused by the frozen angle. The surviving set of frequencies is characterized by Diophantine conditions adapted to the degenerate vector $F=dA,\qquad F_{ij}:=\partial_i A_j-\partial_j A_i.$0.

The physical application is to barriers to transport in three-dimensional incompressible flows. For the swirling Hill’s spherical vortex,

$F=dA,\qquad F_{ij}:=\partial_i A_j-\partial_j A_i.$1

the paper derives a degenerate action–angle–angle description for large swirl $F=dA,\qquad F_{ij}:=\partial_i A_j-\partial_j A_i.$2, studies a time-periodic volume-preserving perturbation, and reports ergodic-partition visualizations at $F=dA,\qquad F_{ij}:=\partial_i A_j-\partial_j A_i.$3 and $F=dA,\qquad F_{ij}:=\partial_i A_j-\partial_j A_i.$4 showing smooth banded structures consistent with surviving KAM tori.

4. Contextual action degeneracy in principal-agent games

In contextual principal-agent games, a principal offers a linear contract $F=dA,\qquad F_{ij}:=\partial_i A_j-\partial_j A_i.$5 and an agent best-responds among available actions. Utilities are

$F=dA,\qquad F_{ij}:=\partial_i A_j-\partial_j A_i.$6

An action is implementable at round $F=dA,\qquad F_{ij}:=\partial_i A_j-\partial_j A_i.$7 under parameter $F=dA,\qquad F_{ij}:=\partial_i A_j-\partial_j A_i.$8 if there exists $F=dA,\qquad F_{ij}:=\partial_i A_j-\partial_j A_i.$9 such that the agent can choose it. It is degenerated if no such contract exists. Relative to a hypothesis set $F_{ij}$0, an action is contextually degenerated if it is not implementable for any $F_{ij}$1:

$F_{ij}$2

A stronger sufficient condition is uniform dominance by another action $F_{ij}$3, meaning

$F_{ij}$4

with strict inequality on a set of positive measure (Feng et al., 21 Oct 2025).

This notion of degeneracy is informational rather than symplectic. Its effect is that informative actions can become unincentivizable across most of the plausible parameter space, so the principal cannot induce or observe them even by varying the contract. The paper identifies this as the mechanism behind a sharp hardness gap between two-action contextual pricing and general principal-agent games with at least three actions. In the three-action, two-dimensional warm-up example, a dominating action $F_{ij}$5 masks the comparison between a purchase action $F_{ij}$6 and the trivial action, preventing the aggressive hypothesis-set shrinkage that is possible in two-action pricing.

The benchmarked performance reflects this obstruction. The paper establishes asymptotically tight pessimistic Stackelberg regret of the form

$F_{ij}$7

up to dimension-dependent constants, and lower bounds

$F_{ij}$8

For the warm-up case $F_{ij}$9 with three actions, the classic Stackelberg regret satisfies

$$\Sigma := \{z\in \Gamma \mid \Delta(z)=0\},\qquad \Delta(z):=\det[F_{ij}(z)].$$0

By contrast, two-action contextual pricing admits near-optimal regret

$$\Sigma := \{z\in \Gamma \mid \Delta(z)=0\},\qquad \Delta(z):=\det[F_{ij}(z)].$$1

The paper therefore describes a double-exponential hardness separation in $$\Sigma := \{z\in \Gamma \mid \Delta(z)=0\},\qquad \Delta(z):=\det[F_{ij}(z)].$$2 once a third action is introduced.

The algorithmic response is optimism with padding and geometry-based hypothesis contraction. The analysis uses an information width

$$\Sigma := \{z\in \Gamma \mid \Delta(z)=0\},\qquad \Delta(z):=\det[F_{ij}(z)].$$3

and a potential

$$\Sigma := \{z\in \Gamma \mid \Delta(z)=0\},\qquad \Delta(z):=\det[F_{ij}(z)].$$4

where $$\Sigma := \{z\in \Gamma \mid \Delta(z)=0\},\qquad \Delta(z):=\det[F_{ij}(z)].$$5 is the $$\Sigma := \{z\in \Gamma \mid \Delta(z)=0\},\qquad \Delta(z):=\det[F_{ij}(z)].$$6-th intrinsic volume. A plausible implication is that, in this setting, degeneracy is best understood as a restriction on what can be learned rather than on what exists.

5. Symmetry-action degeneracy in rational Painlevé V solutions

For a class of rational Painlevé V solutions, action degeneracy arises from affine Weyl-group symmetries. The Hamiltonian form used in the paper is

$$\Sigma := \{z\in \Gamma \mid \Delta(z)=0\},\qquad \Delta(z):=\det[F_{ij}(z)].$$7

with parameters extended to $$\Sigma := \{z\in \Gamma \mid \Delta(z)=0\},\qquad \Delta(z):=\det[F_{ij}(z)].$$8 subject to $$\Sigma := \{z\in \Gamma \mid \Delta(z)=0\},\qquad \Delta(z):=\det[F_{ij}(z)].$$9. The seed solution central to the construction is

$\Sigma$0

denoted $\Sigma$1 (Aratyn et al., 2023).

Only selected symmetry actions are allowed on this seed. Certain simple reflections, specifically $\Sigma$2 and $\Sigma$3, generate divergences and are excluded. The rational orbit is therefore generated by translations $\Sigma$4 and $\Sigma$5:

$\Sigma$6

with parameter lattice

$\Sigma$7

The degeneracy mechanism is explicitly group-theoretic. Two distinct solutions share the same Painlevé V parameters even though the equality between them would require a non-translation Bäcklund transformation that diverges on the seed orbit. The relevant transformations are

$\Sigma$8

When a parameter-matching relation of the form

$\Sigma$9

holds, but the corresponding action on the seed orbit is forbidden by divergence, the result is a two-fold degeneracy: two inequivalent rational solutions of the same equation with identical $F_{ij}$0.

The paper’s central outcome is that this degeneracy occurs for rational solutions generated from the seed $F_{ij}$1 precisely when $F_{ij}$2 is an even integer. Worked examples yield pairs of distinct rational solutions with

$F_{ij}$3

The transformation $F_{ij}$4 flips the sign of $F_{ij}$5, while the action of $F_{ij}$6 changes the $F_{ij}$7 representation without changing the underlying Painlevé V solution. The paper also states that the same formalism extends to higher even-period dressing chains with affine Weyl symmetry $F_{ij}$8.

6. Saddle degeneracy in the gravitational index

In holographic studies of the superconformal index, degeneracy refers to multiple bulk saddles with identical on-shell action and boundary sources, related by a discrete one-form symmetry. In the second-sheet Cardy-like regime,

$F_{ij}$9

the index has asymptotic form

$\dot z^i = F^{ij}(z)\partial_j H(z).$0

up to exponentially small terms. The additive $\dot z^i = F^{ij}(z)\partial_j H(z).$1 term is the signature of saddle degeneracy (Cassani et al., 1 Jun 2026).

The bulk realization uses a discrete two-form gauge theory with BF coupling

$\dot z^i = F^{ij}(z)\partial_j H(z).$2

For the supersymmetric AdS$\dot z^i = F^{ij}(z)\partial_j H(z).$3 black hole saddle, the two-derivative on-shell action is

$\dot z^i = F^{ij}(z)\partial_j H(z).$4

subject to the BPS constraint

$\dot z^i = F^{ij}(z)\partial_j H(z).$5

Degeneracy means that there are several symmetry-related saddles with the same $\dot z^i = F^{ij}(z)\partial_j H(z).$6; it is not a continuum of zero-modes. Summing over them gives

$\dot z^i = F^{ij}(z)\partial_j H(z).$7

At finite volume, the Euclidean time circle caps off in the bulk, producing a cigar-like two-cycle $\dot z^i = F^{ij}(z)\partial_j H(z).$8. A flat two-form holonomy

$\dot z^i = F^{ij}(z)\partial_j H(z).$9

is shifted by the one-form symmetry as

$\Theta_t$00

These shifts label distinct but equivalent saddle sectors. For $\Theta_t$01 holographic SCFTs, $\Theta_t$02, so the correction is $\Theta_t$03. For M5-brane theories on a genus-$\Theta_t$04 Riemann surface, the paper states that the one-form symmetry has order $\Theta_t$05, giving $\Theta_t$06 degenerate saddles.

The same mechanism persists in the gravitational Cardy-like limit, where the boundary decompactifies from $\Theta_t$07 to $\Theta_t$08 and the black hole becomes a supersymmetric black brane. At infinite volume, spontaneous breaking of the one-form symmetry is possible, and the vacuum sector contains $\Theta_t$09 degenerate minima labeled by discrete holonomies. At finite volume, the symmetry is restored by summing over sectors.

7. Comparative perspective

The cited works use “action degeneracy” for technically distinct phenomena, and any unified reading must be qualified. In first-order dynamics, degeneracy is variable rank of $\Theta_t$10 and leads to causal disconnection and superselection sectors (Micheli et al., 2012). In three-dimensional KAM theory, it is the coexistence of one action, two angles, and one zero unperturbed frequency, requiring an intermediate finite transformation block before the infinite KAM step (Vaidya et al., 2014). In contextual search, it is non-implementability over a hypothesis set, causing informational masking and polynomial regret (Feng et al., 21 Oct 2025). In Painlevé V, it is a symmetry-orbit phenomenon in which forbidden Bäcklund actions generate distinct rational solutions with the same equation parameters (Aratyn et al., 2023). In the gravitational index, it is a topological multiplicity of symmetry-related saddles that contributes $\Theta_t$11 to the effective action (Cassani et al., 1 Jun 2026).

This suggests a broad comparative pattern: degeneracy marks a locus where the generic mechanism of continuation fails. The continuation may be dynamical continuation across $\Theta_t$12, KAM continuation of invariant tori under perturbation, exploratory continuation of hypothesis elimination, translational identification along an affine Weyl orbit, or uniqueness of a semiclassical saddle. The consequences are correspondingly different—sectorization, Cantor persistence, hardness of learning, two-fold rational multiplicity, or logarithmic corrections to holographic indices—but in every case the degenerate structure is treated as intrinsic rather than as a removable artifact of coordinates or parametrization.