Arbitrary Evolution Families

• Arbitrary evolution families are two-parameter operator families on Banach spaces that generalize one-parameter semigroups to nonautonomous systems without requiring invertibility.
• They employ time-dependent norms and rate functions to establish general dichotomies, capturing both exponential and polynomial behaviors in dynamic systems.
• They underpin robust stability and admissibility criteria in applications like control theory, differential equations, and holomorphic dynamics, ensuring perturbation tolerance.

An arbitrary evolution family is a two-parameter family of (typically bounded linear) operators that generalizes one-parameter semigroups to nonautonomous settings, without requiring invertibility or time-homogeneity. These families encode the evolution of states in complex, possibly noninvertible, dynamical systems modeled on Banach spaces and appear in diverse mathematical contexts: nonautonomous difference and differential equations, control theory, infinite-dimensional flows, and holomorphic dynamical systems. The theory of arbitrary evolution families enables the extension of dichotomy, stability, and robustness concepts beyond the classical (exponential, invertible) setting, playing a central role in contemporary analysis of time-dependent or structurally irregular systems.

1. Formal Definitions and Setting

Let $(X, \|\cdot\|)$ be a Banach space and $B(X)$ its algebra of bounded linear operators. An evolution family on the half-line is a collection of operators $U = \{ U(t,s) \in B(X) : t \geq s \geq 0 \}$ with the following properties (Dragicevic et al., 2020):

• $U(t, t) = \mathrm{Id}$ for all $t \geq 0$,
• $U(t, s) U(s, r) = U(t, r)$ for all $t \geq s \geq r \geq 0$,
• for each fixed $s \geq 0$ and $x \in X$, both $t \mapsto U(t, s)x$ (for $B(X)$0) and $B(X)$1 (for $B(X)$2) are continuous.

No invertibility or group structure is required. The concept encompasses both invertible (cocycle) and noninvertible cases, and is naturally adapted for nonautonomous ODEs, PDEs, or operator semigroups with time-varying generators. Generalizations include evolution families indexed on arbitrary intervals, in time-dependent domains, or on function spaces of holomorphic mappings (Contreras et al., 2010, Gumenyuk et al., 2022).

2. Rate Functions, Families of Norms, and General Dichotomies

A cornerstone of the modern theory is the abstraction of growth/decay rates via a rate function $B(X)$3, assumed $B(X)$4, strictly increasing, with $B(X)$5 and $B(X)$6 as $B(X)$7. This accommodates exponential, polynomial, and other nonuniform behaviors. The analysis uses a family of equivalent norms $B(X)$8, indexed by time, satisfying

$B(X)$9

for some $U = \{ U(t,s) \in B(X) : t \geq s \geq 0 \}$0, $U = \{ U(t,s) \in B(X) : t \geq s \geq 0 \}$1.

A $U = \{ U(t,s) \in B(X) : t \geq s \geq 0 \}$2–dichotomy (or $U = \{ U(t,s) \in B(X) : t \geq s \geq 0 \}$3–dichotomy) for $U = \{ U(t,s) \in B(X) : t \geq s \geq 0 \}$4 w.r.t. $U = \{ U(t,s) \in B(X) : t \geq s \geq 0 \}$5 is the existence of a family of projections $U = \{ U(t,s) \in B(X) : t \geq s \geq 0 \}$6 satisfying the following (Dragicevic et al., 2020):

• $U = \{ U(t,s) \in B(X) : t \geq s \geq 0 \}$7 for all $U = \{ U(t,s) \in B(X) : t \geq s \geq 0 \}$8,
• on $U = \{ U(t,s) \in B(X) : t \geq s \geq 0 \}$9, $U(t, t) = \mathrm{Id}$0 maps invertibly onto $U(t, t) = \mathrm{Id}$1,
• decay estimates:

$U(t, t) = \mathrm{Id}$2

$U(t, t) = \mathrm{Id}$3

with constants $U(t, t) = \mathrm{Id}$4, and $U(t, t) = \mathrm{Id}$5 interpreted as the inverse on unstable bundles.

The specializations $U(t, t) = \mathrm{Id}$6 and $U(t, t) = \mathrm{Id}$7 recover exponential and polynomial dichotomy, respectively (Dragicevic, 2019). The generalization to arbitrary rates expands the traditional Lyapunov-Perron-Sacker-Sell theory to cover subtle nonuniform and weakly hyperbolic regimes.

3. Admissibility and Characterization Theorems

A major advance is the equivalence between dichotomy and two admissibility conditions, formulated using Banach function spaces. For a closed subspace $U(t, t) = \mathrm{Id}$8 and function spaces $U(t, t) = \mathrm{Id}$9 (absolutely summable inputs), $t \geq 0$0, and $t \geq 0$1, the evolution family $t \geq 0$2 admits a $t \geq 0$3-dichotomy if and only if the following Cauchy problems are uniquely and solvably admissible (Dragicevic et al., 2020):

• For $t \geq 0$4, there is unique $t \geq 0$5 solving

$t \geq 0$6

• For $t \geq 0$7, there is unique $t \geq 0$8 solving

$t \geq 0$9

This extends and unifies decades of admissibility approaches for nonautonomous and weakly hyperbolic systems, establishing that the dichotomy property is not merely about growth splitting, but equivalent to exact controllability of the perturbed system in tailored function spaces. Both admissibility conditions are necessary; omitting one fails to guarantee a genuine dichotomy.

4. Robustness and Stability under Perturbations

Robustness is established at the level of general rate functions and arbitrary evolution families. If $U(t, s) U(s, r) = U(t, r)$0 admits a $U(t, s) U(s, r) = U(t, r)$1–dichotomy with specified constants, then sufficiently small, time-dependent linear perturbations $U(t, s) U(s, r) = U(t, r)$2 preserving suitable decay in $U(t, s) U(s, r) = U(t, r)$3 yield perturbed families $U(t, s) U(s, r) = U(t, r)$4 via Duhamel’s formula:

$U(t, s) U(s, r) = U(t, r)$5

which also admit a $U(t, s) U(s, r) = U(t, r)$6–dichotomy with the same family of time-dependent norms, for all perturbations $U(t, s) U(s, r) = U(t, r)$7 satisfying

$U(t, s) U(s, r) = U(t, r)$8

with $U(t, s) U(s, r) = U(t, r)$9 sufficiently small (Dragicevic et al., 2020). In the exponential case, $t \geq s \geq r \geq 0$0 can decay exponentially; for the polynomial case, a suitable power decay in $t \geq s \geq r \geq 0$1 is required (Dragicevic, 2019). This establishes the persistence of spectral and growth/decay splitting under broad classes of time-dependent or bounded perturbations, substantially generalizing classical perturbation theory results.

5. Specializations: Exponential, Polynomial, and $t \geq s \geq r \geq 0$2-Dichotomies

The arbitrary evolution family framework subsumes several well-studied special cases:

• Exponential dichotomy ($t \geq s \geq r \geq 0$3): The archetypal setting for uniform hyperbolicity and nonuniform hyperbolicity in nonautonomous dynamics.
• Polynomial dichotomy ($t \geq s \geq r \geq 0$4): Enables finer partition of the spectrum associated to “weak hyperbolicity,” where orbits contract/expand at polynomial rates. This extends to “strong nonuniform polynomial dichotomy” in Lyapunov norms (Dragicevic, 2019).
• $t \geq s \geq r \geq 0$5-dichotomy: In the invertible case, for a bijective growth rate $t \geq s \geq r \geq 0$6, the theory further generalizes to “$t \geq s \geq r \geq 0$7–dichotomies,” with splitting estimates in terms of $t \geq s \geq r \geq 0$8. These admit a full characterization via $t \geq s \geq r \geq 0$9–expansiveness and uniform $s \geq 0$0–noncriticality with bounded growth/decay assumptions, unifying the Coppel–Palmer–Elorreaga structure for stability criteria (Dragicevic, 5 Nov 2025).

6. Evolution Families in Complex Analysis and Holomorphic Dynamics

In holomorphic dynamics, evolution families appear as generalizations of semigroups of holomorphic self-maps on complex domains, allowing for inhomogeneous or time-dependent driving (Gumenyuk et al., 2022, Contreras et al., 2010). The Loewner theory and related parametric representation problems hinge on such families:

• Evolution families and reverse evolution families in the unit disk or annulus (including those generated by Bernstein functions and Herglotz vector fields) are described via analogous algebraic identities and absolute continuity conditions.
• The correspondence with semicomplete weak holomorphic vector fields provides a precise infinitesimal description of arbitrary holomorphic evolution families, extending the notion of infinitesimal generators to the nonautonomous, simply/ multiply connected, and noninvertible settings.

7. Illustrative Examples and Applications

Explicit constructions in finite and infinite dimensions show that certain pathologies (such as failure of dichotomy or admissibility) are only avoided when all admissibility conditions are enforced (Dragicevic et al., 2020). General theory is applicable to:

• Nonautonomous ODEs, PDEs with time-varying coefficients,
• Control theory in infinite dimensions,
• Random dynamical systems and ergodic decomposition,
• Evolution equations in complex analysis, including Loewner theory for slits, annuli, and nontrivial multiply connected domains.

Robustness results ensure the practical stability of dichotomy splittings for arbitrary evolution families under small perturbations, vital for numerical analysis and qualitative theory in time-dependent or structurally complex dynamical systems.

---

References:

• “Admissibility and general dichotomies for evolution families” (Dragicevic et al., 2020)
• “Admissibility and polynomial dichotomies for evolution families” (Dragicevic, 2019)
• “h-dichotomies via noncritical uniformity and expansiveness for evolution families” (Dragicevic, 5 Nov 2025)
• “Loewner Theory for Bernstein functions I: evolution families and differential equations” (Gumenyuk et al., 2022)
• “Loewner Theory in annulus I: evolution families and differential equations” (Contreras et al., 2010)