Chaotic Iterations

Updated 29 January 2026

Chaotic iterations are a discrete-time dynamical process on Boolean vectors using strategy-driven, asynchronous updates that meet Devaney’s chaos criteria.
The methodology employs bitwise operations and strong connectivity in the update graph to achieve high unpredictability, making it ideal for robust pseudorandom number generation and hash functions.
Implemented in hardware such as FPGAs, chaotic iterations support secure cryptographic applications and data hiding techniques with proven statistical and performance benchmarks.

Chaotic iterations are a discrete-time dynamical process defined on Boolean vector spaces, wherein the state evolution is governed by a strategy-driven, single-cell asynchronous update mechanism. They serve as a robust mathematical model for generating unpredictability, randomness, and disorder, with concrete proofs—based on topological dynamics—that establish their compliance with Devaney’s chaos criteria. This framework has found rigorous application in pseudorandom number generation, hash functions, data hiding, steganography, and the design of cryptographically secure hardware, particularly in FPGA contexts.

1. Mathematical Framework and Formal Definition

Let $N \geq 2$ be a fixed integer, and let $\mathbb{B}^N$ denote the set of Boolean $N$ -vectors. A strategy sequence $S = (S^1, S^2, \ldots)$ is an infinite sequence with $S^n \in \{1,\ldots,N\}$ . For a chosen iteration function $f: \mathbb{B}^N \rightarrow \mathbb{B}^N$ and initial state $x^0 \in \mathbb{B}^N$ , chaotic iterations are defined as:

$x_i^n = \begin{cases} x_i^{n-1} & \text{if}\; i \neq S^n, \ f(x^{n-1})_i & \text{if}\; i = S^n. \end{cases}$

for $n \geq 1$ and $i = 1,\ldots,N$ .

This can be compactly expressed in vector form: $x^n = x^{n-1} \oplus e_{S^n} \cdot (f(x^{n-1}) - x^{n-1}),$ where $e_j$ is the unit vector at position $j$ (Bahi et al., 2016).

The phase space is $\mathbb{X} = \mathbb{S} \times \mathbb{B}^N$ with metric

$d((S,x),(S',x')) = \sum_{i=1}^N |x_i - x'_i| + \sum_{n=1}^\infty \frac{|S^n-S'^n|}{2^n}.$

The core dynamical operator is

$G_f(S, x) = (\sigma(S), F_f(S^1, x)),$

with $\sigma$ the left-shift on strategies, and $F_f(j, x)$ updating only component $j$ by $f_j(x)$ .

2. Topological Chaos: Devaney’s Criteria

Devaney’s definition of chaos in a metric space encompasses three key conditions: (1) topological transitivity, (2) density of periodic points (regularity), and (3) sensitivity to initial conditions. Guyeux and Bahi establish that chaotic iterations realize Devaney’s chaos when the asynchronous iteration graph $\Gamma(f)$ —with vertices the Boolean states and arcs $x \to F_f(i,x)$ for all $x$ and $i$ —is strongly connected (Guyeux et al., 2015, Bahi et al., 2011, Guyeux et al., 2016):

Transitivity: For any pair of open sets, an orbit segment can be constructed to connect them, leveraging the connectivity of $\Gamma(f)$ .
Density of Periodic Points: Arbitrarily close periodic orbits can be constructed by cycling finite strategy segments.
Sensitivity: As proved by Banks’ theorem, the combination of transitivity and density implies sensitivity; small perturbations in the initial state or strategy are magnified unpredictably in future iterates.

Expansivity and infinitely large topological entropy can also be shown for CI with vectorial negation as the iteration function, strengthening unpredictability (Guyeux et al., 2016).

3. Algorithmic Construction and Variants

Typical iteration functions employ bitwise negation: $f_0(x_1, ..., x_N) = (\neg x_1, ..., \neg x_N).$ Generalizations—such as XOR-based CI—use

$x^{n+1} = x^n \oplus s^n,$

where $s^n$ is a unit vector with $1$ at position $S^n$ (Bahi et al., 2016).

Pseudorandom strategies $S$ may be generated by modulo-reducing outputs of a secondary PRNG. Version 2 CI (“irregular decimation”) avoids revisiting the same cell within one round, maximizing mixing (Bahi et al., 2016, Bahi et al., 2017).

High-level pseudocode for Version 1 CI PRNG (Bahi et al., 2016):

def chaotic_iteration_PRNG(x, PRNG1, PRNG2, c):
    a = PRNG1()
    m = (a % 2) + c
    for i in range(m):
        b = PRNG2()
        S = b % N
        x[S] = not x[S]  # flip S-th bit
    return x

4. Hardware Implementation and FPGA Mapping

Chaotic iterations are inherently compatible with digital hardware due to their reliance on integer and bit-wise operations. In FPGA contexts, the CI core is mapped to parallelized bit-flip and XOR logic, with PRNG modules (e.g., two 64-bit XORshifts and a compact BBS) serving as strategy and mixing generators (Bahi et al., 2016, Bakiri et al., 2016).

On an Altera Cyclone II EP2C8Q208C8 device, CI PRNG achieves throughput exceeding 6000 Mbit/s. The architecture employs independent PRNG cores, a CI block for masked updates, and registers for CI state. Hardware resources are efficiently used: less than 75% of available logic elements for full pipeline (Bahi et al., 2016, Bakiri et al., 2016).

Empirical performance:

Design	Logic Elements	Throughput (Gb/s)	Statistical Robustness
XORshift-only	<1000	6.9	Fails BigCrush
CI-PRNG (CIPRNG)	~7000	>6.0	Passes BigCrush
BBS-only (small)	~1200	0.5	Fails NIST

5. Security Properties: Randomness and Cryptographic Suitability

CI-based generative models have been subjected to extensive statistical testing:

NIST SP 800-22 Battery: CI-PRNG passes 15/15 tests, outperforming XORshift (14/15) and small-modulus BBS (2/15). P-values for core tests (monobit, block frequency, runs, cumulative sums) range from 0.073 to 0.819, well within accepted bounds (Bahi et al., 2016).
TestU01/DieHard: CI mixing elevates linear PRNGs to full compliance with BigCrush and DieHard, which pure linear PRNGs often fail (Bakiri et al., 2016, Bahi et al., 2017).

CI generators exhibit high sensitivity to seed and strategy variations (guaranteed by Devaney’s chaoticity), strong diffusion and confusion properties, and cryptographic unpredictability. When underlying graphs are doubly stochastic, uniform output distribution is obtained (Bahi et al., 2011).

6. Applications: Information Hiding, Cryptography, Hash Functions

Chaotic iterations serve as foundational mechanisms for data hiding and steganography. In watermarking, a CI-PRNG is used to both encrypt and pseudorandomly distribute watermark bits in the cover image, ensuring robustness and stego-security under the watermark-only attack model (Friot et al., 2011, Guyeux et al., 2010).

In hash constructions, CI acts as a post-processing stage over classical hashes or as a full compressor:

Hash Function Example: Given input $M$ , preprocess to seed $E$ , derive strategy $S$ via rotated blocks, and iterate with vectorial negation (Guyeux et al., 2015, Guyeux et al., 2016, Bahi et al., 2017). Output digest is the final CI state, typically in hexadecimal form.
Security Analysis: Avalanche effect ensures $>50\%$ bit change for single input bit flip, with confusion/diffusion confirmed through empirical avalanche metrics ( $\bar{B} \approx N/2$ changed bits, $P \approx 50\%$ mean changed probability) (Lin et al., 2017, Bahi et al., 2011).

Symmetric cryptographic modes (e.g., CBC) can be rigorously recast as CI systems, with chaos guaranteed under strong connectivity of the update graph (Fang et al., 2016).

7. Extensions: Neural Networks and Generalized CI

CI formalism generalizes to strategies selecting subsets of cells at each iteration, and has been directly embedded in recurrent neural network architectures. Bahi et al. prove equivalence between CI dynamics and certain global multi-layer perceptrons: if the underlying iteration graph is strongly connected, the neural model is chaotic in Devaney’s sense (Bahi et al., 2016, Bahi et al., 2011). Feedforward MLPs trained on CI-generated data display poor predictive performance, reflecting the inherent unpredictability and resistance to learning typical of chaotic systems.

Chaotic iterations provide a rigorously validated mathematical infrastructure for discrete chaos, supporting high-performance, cryptographically suitable PRNGs, hash functions, and robust data hiding schemes, with direct digital hardware compatibility and broad, theoretically grounded security guarantees.