Componentwise Timeshift Transformation (CTT)

• CTT is a dynamical-systems equivalence that assigns independent temporal shifts to network nodes, reducing delay parameter redundancy to the cycle space dimension.
• CTT preserves invariant dynamics, spectral properties, and stability features such as equilibria, periodic orbits, and Lyapunov exponents under time shifts.
• CTT enables classification of delay-coupled networks into equivalence classes based on cycle-sum invariants, facilitating model reduction and comparative analyses.

The Componentwise Timeshift Transformation (CTT) is a dynamical-systems equivalence that enables the classification of delay-coupled networks according to their essential dynamical properties. CTT operates by assigning an independent temporal shift to each node in a network, thereby transforming both the state variables and the set of connection delays. Critically, CTT reveals that the observable dynamics, spectral properties, and stability characteristics of such networks depend not on the raw delays assigned to each connection, but rather on cycle-sum invariants determined by the underlying network topology. Consequently, the effective dimension of the parameter space of delays is generically not greater than the cycle space dimension of the network’s graph, providing a canonical form for delay-mediated dynamics (Lücken et al., 2013).

1. Mathematical Definition of CTT

Consider a network with $N$ nodes, $x_j(t)\in\mathbb{R}$, $j=1,...,N$, described by the (multi-)digraph $(\mathcal{N}, \mathcal{E})$ where $\mathcal{N} = \{1,\dots,N\}$ and $$\mathcal{E} \subseteq \mathcal{N}\times \mathcal{N}$$. Each directed edge $\ell \in \mathcal{E}$ has a source $s(\ell)$, target $t(\ell)$, and delay $\tau(\ell)\geq0$. The network dynamics is governed by the delay differential equations: $x_j(t)\in\mathbb{R}$0 where $x_j(t)\in\mathbb{R}$1 is the set of incoming edges at node $x_j(t)\in\mathbb{R}$2.

The CTT introduces a family of temporal offsets $x_j(t)\in\mathbb{R}$3 and defines new variables as

$x_j(t)\in\mathbb{R}$4

Under this transformation, the network retains its graph structure, but the delays are shifted to

$x_j(t)\in\mathbb{R}$5

yielding the transformed system: $x_j(t)\in\mathbb{R}$6 CTT is formalized on semiflows over Banach spaces of histories and involves a timeshift operator $x_j(t)\in\mathbb{R}$7 mapping histories $x_j(t)\in\mathbb{R}$8 to their appropriately shifted versions in $x_j(t)\in\mathbb{R}$9. The rigorous commutative relations $j=1,...,N$0 (for the original semiflow $j=1,...,N$1 and the transformed semiflow $j=1,...,N$2) guarantee that CTT is a true dynamical equivalence.

2. Dynamical Equivalence and Invariance Theorems

CTT guarantees strong equivalence between the semiflows $j=1,...,N$3 and $j=1,...,N$4. The following invariances under CTT have been established:

• Spectrum Invariance: The characteristic exponents (e.g., those of equilibria and periodic orbits) are identical for any CTT-related realizations. Specifically, for an equilibrium $j=1,...,N$5 of the original system, the characteristic exponents in the original and transformed systems coincide. For $j=1,...,N$6-periodic solutions $j=1,...,N$7, Floquet exponents are preserved.
• Invariant Sets Correspondence: Positively invariant sets, fully invariant sets, strongly invariant hulls, and minimal invariant sets are carried into each other by the timeshift operator $j=1,...,N$8. There is a bijective correspondence between periodic orbits, equilibria, connecting orbits, and other invariant sets.
• Lyapunov Exponents and Stability: Lipschitz continuity of $j=1,...,N$9 and its inverse ensures that corresponding invariant sets have identical maximal Lyapunov exponents and equivalent stability properties, including asymptotic stability.

These results hold for a wide class of semiflows, including those governed by delay differential equations (DDEs) under standard Lipschitz conditions (Lücken et al., 2013).

3. Invariance Properties and Qualitative Dynamics

The preservation of dynamical features under CTT extends to all core qualitative characteristics:

• Qualitative Dynamics: All attractors, their (in)stability, invariant manifolds, and their qualitative features are invariant under the transformation.
• Types of Invariant Sets: Strong, positive, and full invariance are preserved.
• Bifurcation and Spectral Data: The bifurcation structure with respect to delay parameters is unaffected so long as parameter representatives remain in the same CTT-class.

This invariance ensures that the classification and analysis of delay-coupled networks can proceed on the basis of CTT-classes rather than individual (and often redundant) sets of delay assignments.

4. Classification of Delay-Parameter Equivalence Classes

At the level of delay parameters, two assignments $(\mathcal{N}, \mathcal{E})$0 and $(\mathcal{N}, \mathcal{E})$1 are related by CTT if and only if the following holds for all undirected cycles $(\mathcal{N}, \mathcal{E})$2: $(\mathcal{N}, \mathcal{E})$3 where $(\mathcal{N}, \mathcal{E})$4 encodes the direction of $(\mathcal{N}, \mathcal{E})$5 in the cycle. Thus, all cycle (semi-cycle) roundtrip sums of delays,

$(\mathcal{N}, \mathcal{E})$6

are invariants under CTT.

Importantly, the number of independent invariants is dictated by the cycle space dimension $(\mathcal{N}, \mathcal{E})$7, where $(\mathcal{N}, \mathcal{E})$8 and $(\mathcal{N}, \mathcal{E})$9 the number of nodes. By making appropriate timeshift choices (for example, setting delays along a spanning tree to zero), it is possible to reduce the parameter space of effective delays to exactly $\mathcal{N} = \{1,\dots,N\}$0 essential parameters, corresponding to the independent cycles of the connectivity graph.

Graph Size	Links $\mathcal{N} = \{1,\dots,N\}$1	Nodes $\mathcal{N} = \{1,\dots,N\}$2	Cycle Space Dimension $\mathcal{N} = \{1,\dots,N\}$3
Unidirectional ring	$\mathcal{N} = \{1,\dots,N\}$4	$\mathcal{N} = \{1,\dots,N\}$5	$\mathcal{N} = \{1,\dots,N\}$6
Bidirectional edge	$\mathcal{N} = \{1,\dots,N\}$7	$\mathcal{N} = \{1,\dots,N\}$8	$\mathcal{N} = \{1,\dots,N\}$9
General digraph	$$\mathcal{E} \subseteq \mathcal{N}\times \mathcal{N}$$0	$$\mathcal{E} \subseteq \mathcal{N}\times \mathcal{N}$$1	$$\mathcal{E} \subseteq \mathcal{N}\times \mathcal{N}$$2

This reduction is generically maximal; fewer than $$\mathcal{E} \subseteq \mathcal{N}\times \mathcal{N}$$3 parameters are possible only via additional linear relations among the delays, which are non-generic (positive codimension).

5. Illustrative Examples and Canonical Forms

The application of CTT to concrete networks underscores its classification power:

• Unidirectional Ring of $$\mathcal{E} \subseteq \mathcal{N}\times \mathcal{N}$$4 Nodes: With dynamics

$$\mathcal{E} \subseteq \mathcal{N}\times \mathcal{N}$$5

the Baldi–Atiya timeshift, $$\mathcal{E} \subseteq \mathcal{N}\times \mathcal{N}$$6, yields all but one delay identically zero:

$$\mathcal{E} \subseteq \mathcal{N}\times \mathcal{N}$$7

Alternatively, the Perlikowski–Popovych homogenization,

$$\mathcal{E} \subseteq \mathcal{N}\times \mathcal{N}$$8

transforms all delays to be equal: $$\mathcal{E} \subseteq \mathcal{N}\times \mathcal{N}$$9.

• Bidirectional Two-Node Ring: For edge delays $\ell \in \mathcal{E}$0, setting $\ell \in \mathcal{E}$1, $\ell \in \mathcal{E}$2 reduces both delays to $\ell \in \mathcal{E}$3. Thus, the two-parameter family collapses to a single effective delay.

These examples concretely demonstrate the reduction in delay parameters and the invariance of system trajectories (modulo componentwise shifts).

6. Conclusion and Impact on Delay-Differential Network Theory

CTT demonstrates that for networks of delay-coupled systems, only $\ell \in \mathcal{E}$4 independent delay parameters govern all qualitative and spectral dynamical features. This revision of the effective parameter space has profound implications for the modeling, reduction, and comparative study of DDE-networks and related systems, revealing hidden symmetries and drastically reducing model redundancy. Hence, researchers can classify dynamics up to CTT-equivalence, focusing on cycle-sum data rather than arbitrary sets of connection delays (Lücken et al., 2013).