Generalized First Integrals in Dynamical Systems

• Generalized first integrals are extended conservation laws for dynamical systems, generalizing classical invariants to non-metric and computational frameworks.
• They include rational, polynomial, and monomial forms derived via algebraic, resonance, and symmetry methods that enhance integrability analysis.
• They are pivotal in algorithmic computation and geometric numerical methods, fostering new approaches in dynamical system analysis.

Generalized first integrals are conserved quantities of differential or difference systems obtained by enlarging the classical requirement that a function remain constant along solution curves. In the current literature, the adjective generalized does not denote a single uniform definition; instead, it labels several extensions of the classical notion, including generalized rational quotients $G/H$, higher-order polynomials in velocities, monomial and polynomial invariants organized by resonance, approximate adiabatic invariants, and discrete or Lie-group analogues of conservation laws. This body of work suggests that generalized first integrals are best understood as a family of conservation-law formalisms adapted to non-metric geometries, non-variational equations, resonant normal forms, and computational settings (Cong et al., 2014, Mitsopoulos et al., 2023, Avendaño-Camacho et al., 2013).

1. Definitions and terminological scope

A first integral of a dynamical system generated by a vector field $X$ is a smooth function $F$ such that $\mathcal{L}_X F=0$; equivalently, for a second-order system, a function $I(t,q,\dot q)$ satisfying $dI/dt=0$ along solutions. This basic definition underlies several generalized variants. For analytic differential systems $\dot x=f(x)$, a generalized rational first integral is a function

$F(x)=\frac{G(x)}{H(x)}$

with $G$ and $H$ analytic near the origin and $X$0. For holonomic autonomous systems

$X$1

a higher-order first integral is taken to be a polynomial of degree $X$2 in the velocities, with totally symmetric tensor coefficients. For diagonal linear systems, monomial first integrals have the form $X$3 and satisfy a resonance condition involving the eigenvalues of the linear part (Tudoran, 2014, Cong et al., 2014, Mitsopoulos et al., 2023, Grašič et al., 30 Jul 2025).

The literature also distinguishes generalized first integrals from objects called generalized integrals in operator theory and special-function analysis. In that separate usage, a generalized integral is a linear functional extending the standard integral to functions with finitely many homogeneous non-integrable terms at the endpoints; it is used for Macdonald and Gegenbauer functions and for Green functions with point interactions, rather than for conserved quantities of dynamical flows (Dereziński et al., 2023).

2. Algebraic, resonant, and invariant-theoretic formulations

For autonomous systems

$X$4

with $X$5 diagonal and $X$6 containing no constant or linear terms, monomial first integrals of the linear part are characterized by

$X$7

This gives the affine monoid

$X$8

whose minimal generating set is the Hilbert basis $X$9. The corresponding algebra $F$0 of polynomial first integrals is a finitely generated $F$1-algebra, and the same computational framework extends to algebraic complex eigenvalues by embedding the eigenvalues into a number field, forming an integer matrix $F$2, computing $F$3, intersecting with $F$4, and extracting a Hilbert basis with Gröbner-basis methods. The same Diophantine and commutative-algebraic machinery describes polynomial invariants in parameter space and resonant monomials for Poincaré-Dulac normal forms, with

$F$5

governing resonant terms in the $F$6-th component (Grašič et al., 30 Jul 2025).

A complementary resonance theory applies to generalized rational first integrals of analytic systems. If $F$7 and $F$8 has eigenvalues $F$9, then the maximal number of functionally independent generalized rational first integrals near the equilibrium is at most the dimension of the minimal subspace of $\mathcal{L}_X F=0$0 containing

$\mathcal{L}_X F=0$1

Analogous bounds are stated for semi-quasi-homogeneous systems via Kowalevskaya exponents, for neighborhoods of periodic orbits via multipliers $\mathcal{L}_X F=0$2, and for periodic differential systems via Floquet multipliers $\mathcal{L}_X F=0$3. A key lemma states that functional independence of generalized rational functions implies functional independence of their lowest-order rational homogeneous terms, which makes resonance counting decisive for necessary integrability conditions (Cong et al., 2014).

For planar polynomial vector fields, the algebraic-differential hierarchy

$\mathcal{L}_X F=0$4

is made algorithmic through generalized extactic curves. In this setting, vanishing of the appropriate extactic determinant for a degree bound $\mathcal{L}_X F=0$5 is equivalent to the existence of a first integral of the corresponding class, and the output is a defining differential equation from which the integral can be reconstructed (Chèze et al., 2017).

3. Geometric constructions from connections, Killing tensors, and Hamiltonian extensions

For autonomous holonomic systems with symmetric, possibly non-metrical connection,

$\mathcal{L}_X F=0$6

higher-order first integrals are sought in the polynomial ansatz

$\mathcal{L}_X F=0$7

with totally symmetric coefficients. Substituting into $\mathcal{L}_X F=0$8 yields a system of PDEs that splits into a geometric part, determined by the connection, and a dynamical part, involving the generalized forces. The leading coefficient must satisfy the generalized Killing tensor equation

$\mathcal{L}_X F=0$9

and lower-rank coefficients are obtained recursively. In Riemannian cases these tensors reduce to ordinary Killing tensors, but the construction remains valid for arbitrary symmetric connections and therefore produces non-Noetherian first integrals in non-metrical settings (Mitsopoulos et al., 2023).

A parallel formulation states that autonomous and time-dependent first integrals of any order can be written systematically as either polynomials in $I(t,q,\dot q)$0 with tensor coefficients or as factorizable expressions with exponential time dependence. In this formulation, first integrals of order $I(t,q,\dot q)$1 correspond to generalized Killing tensors of rank $I(t,q,\dot q)$2 together with recursive relations involving $I(t,q,\dot q)$3. The explicit low-order cases recover generalized Killing vectors for linear integrals, rank-two generalized Killing tensors for quadratic integrals, and rank-three generalized Killing tensors for cubic integrals (Mitsopoulos et al., 2023).

Another geometric generalization appears for natural Hamiltonians extended by one degree of freedom,

$I(t,q,\dot q)$4

If $I(t,q,\dot q)$5 satisfies

$I(t,q,\dot q)$6

then $I(t,q,\dot q)$7 is a new first integral of $I(t,q,\dot q)$8, independent of $I(t,q,\dot q)$9. The existence of these integrals is tied to the geometry of the configuration manifold $dI/dt=0$0: when $dI/dt=0$1 depends only on positions, the maximal number of linearly independent solutions exists if and only if $dI/dt=0$2 has constant curvature, while for momentum-polynomial $dI/dt=0$3 the leading tensor must satisfy the self-conformal Killing condition

$dI/dt=0$4

The same construction extends to Poisson manifolds, and for quadratic first integrals the Laplace-Beltrami quantization satisfies $dI/dt=0$5 if and only if $dI/dt=0$6 (Chanu et al., 2011).

4. Symmetries, Poisson structures, and exact integrability mechanisms

A direct route from symmetry data to first integrals is available when a vector field $dI/dt=0$7 admits linearly independent infinitesimal symmetries $dI/dt=0$8 satisfying $dI/dt=0$9 and

$\dot x=f(x)$0

Under these assumptions, each structure coefficient $\dot x=f(x)$1 is a first integral, and so is every Lie derivative $\dot x=f(x)$2. When $\dot x=f(x)$3, the bivector

$\dot x=f(x)$4

defines a rank-two Poisson structure, the associated Casimirs are precisely the functions annihilated by both $\dot x=f(x)$5 and $\dot x=f(x)$6, and the symplectic leaves are the two-dimensional integral manifolds tangent to $\dot x=f(x)$7. If there exists $\dot x=f(x)$8 such that

$\dot x=f(x)$9

then $F(x)=\frac{G(x)}{H(x)}$0 is Hamiltonian with respect to $F(x)=\frac{G(x)}{H(x)}$1 (Tudoran, 2014).

For the differential chains generated by

$F(x)=\frac{G(x)}{H(x)}$2

including the Riccati chain ($F(x)=\frac{G(x)}{H(x)}$3) and the Abel chain ($F(x)=\frac{G(x)}{H(x)}$4), the determination of $F(x)=\frac{G(x)}{H(x)}$5 generalized symmetries of the $F(x)=\frac{G(x)}{H(x)}$6-th order equation produces $F(x)=\frac{G(x)}{H(x)}$7 functionally independent first integrals without integration. If $F(x)=\frac{G(x)}{H(x)}$8 and $F(x)=\frac{G(x)}{H(x)}$9 satisfy the defining relation for the symmetry, then $G$0 is a first integral; in the explicit construction,

$G$1

The remaining integral follows from the Jacobi last multiplier

$G$2

which reduces the problem to a Bernoulli auxiliary equation and yields an explicit general solution formula for the chain (Muriel et al., 2021).

Central-force dynamics in $G$3 dimensions supplies a further exact integrability scheme derived without Noether’s theorem or dynamical symmetries. By solving the first-integral determining equation in polar variables through the method of characteristics, one obtains a complete set of $G$4 functionally independent first integrals, consisting of energy, angular momentum, a generalized Laplace-Runge-Lenz vector, and a temporal quantity involving $G$5 explicitly. The generalized Laplace-Runge-Lenz vector reduces to the standard one for the inverse-square force and becomes multi-valued for precessing bounded trajectories (Anco et al., 2015).

A specialized extension occurs for generalized Darboux-Halphen systems. For systems with a common additive term, the classical conserved quantity

$G$6

remains valid. For systems with individual additive terms, conserved quantities are obtained by similarity transformations that relate the generalized system to one with known integrals (Chanda et al., 2016).

5. Algorithmic and symbolic computation

The modern theory of generalized first integrals is strongly algorithmic. For planar polynomial vector fields, generalized extactic curves convert the search for rational, Darbouxian, Liouvillian, and Riccati first integrals with bounded degree into kernel computations for structured linear maps. The probabilistic algorithm has arithmetic complexity

$G$7

where $G$8 is the degree bound and $G$9 is the exponent of linear algebra; the deterministic variant has complexity $H$0 and performs $H$1 univariate polynomial factorizations. This replaces earlier recombination-heavy approaches by a unified linear-algebraic framework (Chèze et al., 2017).

For polynomial ODE systems, the integrating factor matrix method seeks a skew-symmetric matrix $H$2 satisfying

$H$3

together with a curl-free condition on $H$4. In dimension two this reduces to a scalar integrating factor $H$5; in dimension three one uses parameterized skew-symmetric $H$6 ansätze. Applied to Lotka-Volterra systems with constant terms, the method yields explicit parameter conditions for polynomial, logarithmic, and power-type first integrals, and it reproduces known integrals while also identifying new families (Saputra et al., 2010).

For ordinary difference equations beyond Lagrangian methods, the adjoint-equation method is organized around the discrete identity

$H$7

If a chosen substitution makes the adjoint equation vanish on solutions, then $H$8 and $H$9 is a first integral. The method does not require a variational formulation, admits generalized substitutions

$X$00

and was applied to invariant mappings and discretizations of second- and third-order ODEs (Dorodnitsyn et al., 2013).

6. Approximate invariants and geometric numerical preservation

Generalized first integrals are not restricted to exact invariants. In generalized slow-fast Hamiltonian systems on a product symplectic manifold with

$X$01

an approximate first integral of order $X$02 is a formal series

$X$03

such that $X$04. Under a periodicity hypothesis for the unperturbed flow, an $X$05-action, a momentum map $X$06, and the adiabatic condition $X$07, one obtains explicit global formulas for a second-order approximate first integral,

$X$08

with $X$09 and $X$10 expressed through the averaging operator $X$11 and the integrating operator $X$12. The method is coordinate-free and applies, among other examples, to the elastic pendulum and a charged particle in a slowly varying magnetic field (Avendaño-Camacho et al., 2013).

A different extension concerns numerical time-stepping on Lie groups and homogeneous manifolds. The discrete gradient approach is generalized by introducing a trivialized discrete differential $X$13 satisfying

$X$14

With a skew-symmetric discrete bivector $X$15, the one-step method

$X$16

preserves $X$17 exactly, because the increment of $X$18 is the value of a skew-symmetric form on two identical arguments. Symmetric midpoint-type and averaged-vector-field-type constructions are given, and the framework extends to higher even order through collocation while preserving the first integral exactly at each step (Celledoni et al., 2013).

These developments suggest that the modern concept of generalized first integrals spans exact algebraic invariants, geometric and non-Noetherian constants of motion, asymptotic adiabatic invariants, and discretely preserved quantities. The unifying theme is not a single formal definition but the systematic enlargement of conservation laws to settings in which classical integrability criteria are too restrictive.