Idempotent Boolean Functions

• Idempotent Boolean functions are defined by their invariance under the Frobenius automorphism, linking field theory with rotation-symmetric functions.
• They are efficiently parametrized using orbit-based encoding, compressing the representation from 2^(2^n) to 2^N for practical enumeration and search methods.
• Their clone-theoretic structure and optimal nonlinearity properties in moderate dimensions make them crucial for cryptographic design and finite lattice operations.

Idempotent Boolean functions constitute a mathematically distinguished subclass of Boolean functions, characterized by invariance under the Frobenius automorphism $x \mapsto x^2$ over finite fields of characteristic $2$. Their algebraic structure links them to rotation-symmetric functions in certain bases, and they feature prominently in both clone theory for finite lattices and cryptographic design methodology due to their inherent algebraic constraints and symmetry properties (Carlet et al., 31 Jan 2026, Botur et al., 2018).

1. Algebraic Characterizations

Let $\F_2 = \{0,1\}$ denote the two-element field and $\F_{2^n}$ its extension field of order $2^n$. Boolean functions can be represented in two fundamental forms:

• Univariate (field-theoretic): $f : \F_{2^n} \to \F_2$
• Multivariate (vector): $F : \F_2^n \to \F_2$, identifying $\F_{2^n} \cong \F_2^n$ via a chosen basis

A function $f$ is idempotent if

$f(x^2) = f(x) \quad \forall x \in \F_{2^n}.$

The equivalent characterization in coordinate form involves a fixed $\F_2$–linear map $S_n : \F_2^n \to \F_2^n$ (squaring matrix in a polynomial basis), such that

$F(S_n(X)) = F(X) \quad \forall X \in \F_2^n.$

Alternatively, in a normal basis, the squaring operation is a cyclic rotation (left shift), so idempotence is equivalent to rotation symmetry: $F(\rho(X)) = F(X),$ with $\rho$ as the coordinate left shift. Thus, the set of idempotent Boolean functions coincides with the set of Frobenius-invariant functions, and, in certain bases, with rotation-symmetric functions (Carlet et al., 31 Jan 2026).

2. Orbit Structure and Parametrization

The Frobenius map partitions $\F_{2^n}$ into orbits: $$\mathcal{O}(x) = \{x, x^2, x^{2^2}, \dots, x^{2^{n-1}}\},$$ with $x^{2^n} = x$ ensuring finiteness. By idempotence, $f$ must be constant on each orbit. The total number of distinct orbits is

$N = \frac{1}{n} \sum_{k=0}^{n-1} 2^{\gcd(n, k)}.$

Each idempotent Boolean function is fully determined by its values on orbits, i.e., a bit vector of length $N$. This yields a compression of the representation compared to the general case of $2^{2^n}$ functions, reducing the space to $2^N$ functions (Carlet et al., 31 Jan 2026).

3. Clone-Theoretic Structure

Clone theory formalizes operations closed under composition and containing all projections. On a general finite lattice $L = (L, \wedge, \vee, 0, 1)$, an idempotent aggregation function $f: L^n \to L$ satisfies:

• Monotonicity: $f$ is monotone in each coordinate,
• Boundary: $f(0,\dots,0)=0$ and $f(1,\dots,1)=1$,
• Diagonal idempotence: $f(x,\dots,x) = x$.

The set $\text{Id}(L)$ of all such functions forms a clone and is generated by $\wedge$, $\vee$, and a family of ternary operations $\ell_{(a,b,c),d}$ defined by

$$\ell_{(a,b,c),d}(x,y,z) = \begin{cases} d \wedge (x \vee y \vee z), &\text{if } x \leq a, y \leq b, z \leq c\ x \vee y \vee z, &\text{otherwise} \end{cases}.$$

In the Boolean lattice $L = \{0,1\}$, these operations reduce to just $\wedge, \vee$ (the monotone clone $M_2$), as all ternary generators can be expressed using them (Botur et al., 2018).

4. Algorithmic Construction: Orbit-Based Evolution

Evolutionary algorithms (EAs) have been employed to search for highly nonlinear idempotent Boolean functions, especially in cryptographic design. Two principal encoding strategies are used:

• Full truth table: Bitstring of length $2^n$; enforcing idempotence requires penalizing non-conforming functions during search.
• Orbit-based encoding: Bitstring of length $N$ (number of orbits); any genome is valid by construction.

Standard EA operators (crossover, mutation) disrupt the required invariance in the full truth table setting, resulting in performance degradation unless strong repair mechanisms are used. Orbit-based encoding, by contrast, ensures idempotence inherently and allows efficient use of generic search operators. Empirically, this restriction sharply reduces genome size and leads to optimal or near-optimal nonlinearity for $n \leq 8$, with a small gap for larger $n$ due to increased search difficulty (Carlet et al., 31 Jan 2026).

Pseudocode Excerpts

Orbit enumeration:

def ComputeOrbits(n, p_n): visited = set() orbits = [] for x in F_{2^n}: if x not in visited: O = [] y = x while True: O.append(y) visited.add(y) y = y**2 if y == x: break orbits.append(O) return orbits

Decoding genotypes to truth tables:

def DecodeGenome(g, orbits): F = {} for i, O in enumerate(orbits): for x in O: F[x] = g[i] return F

(Carlet et al., 31 Jan 2026)

5. Enumeration and Generating Sets

On the Boolean lattice, the idempotent Boolean functions correspond precisely to the monotone Boolean functions nonconstant at both boundaries. Their number for $n$ variables is $D(n) - 2$, with $D(n)$ the $n$-th Dedekind number:

• $n=1$: $1$
• $n=2$: $4$
• $n=3$: $18$
• $n=4$: $166$
• $n=5$: $7\,579$
• $n=6$: $7\,828\,350$ (OEIS A000372).

Constructionally, all idempotent Boolean functions are generated by $\wedge$ and $\vee$. For more general lattices, the inclusion of special ternary operations is essential for clone generation (Botur et al., 2018).

Setting	Generating Set	Cardinality
Boolean lattice $\{0,1\}$	$\wedge, \vee$	$D(n)-2$ functions
General finite lattice $L$	$\wedge, \vee, \ell_{(a,b,c),d}$	Finitary, via ternary generators

6. Cryptographic Properties and Nonlinearity

Nonlinearity is a core criterion for cryptographic Boolean function strength. For idempotent classes, the maximal nonlinearity for $n \leq 8$ matches the general upper bound (bent functions exist); for $n > 8$, a gap of a few units emerges:

• $n=5$ ($N=8$): $12$ (optimal)
• $n=6$ ($N=14$): $28$ (bent, optimal)
• $n=7$ ($N=20$): $56$ (optimal)
• $n=8$ ($N=36$): $120$ (bent, optimal)
• $n=9$ ($N=60$): $240$ vs. bound $242$
• $n=12$ ($N=352$): $1992$ vs. bound $2016$

This suggests that for moderate $n$, the constraints of idempotence do not prohibit achieving cryptographically significant nonlinearity, but the evolutionary search faces growing combinatorial obstructions as $n$ increases (Carlet et al., 31 Jan 2026).

7. Structural and Theoretical Insights

The study of idempotent Boolean functions reveals several salient theoretical features:

• Their algebraic invariance links field-theoretic and combinatorial symmetries, unifying rotation-symmetry and Frobenius-fixedness via basis choice.
• In the clone-theoretic setting, the idempotent Boolean functions exactly coincide with the monotone clone on the Boolean lattice, forming the generator set for all monotone operations.
• The Frobenius-orbit framework enables compact parametrization, essential for both theoretical enumeration and efficient evolutionary search.

A plausible implication is that algorithmic advances exploiting orbit structure, or hybridizing clone-theoretic generative steps with search-based paradigms, are necessary for scaling the construction of highly nonlinear idempotent functions beyond the smallest dimensions (Carlet et al., 31 Jan 2026, Botur et al., 2018).