Transposition of variables is hard to describe

The function $p_{xy}$ that interchanges two logical variables $x,y$ in formulas is hard to describe in the following sense. Let $F$ denote the Lindenbaum-Tarski formula-algebra of a finite-variable first order logic, endowed with $p_{xy}$ as a unary function. Each equational axiom system for the equational theory of $F$ has to contain, for each finite $n$, an equation that contains together with $p_{xy}$ at least $n$ algebraic variables, and each of the operations $\exists, =, \lor$. This solves a problem raised by Johnson [J. Symb. Logic] more than 50 years ago: the class of representable polyadic equality algebras of a finite dimension $α\ge 3$ cannot be axiomatized by adding finitely many equations to the equational theory of representable cylindric algebras of dimension $α$. Consequences for proof systems of finite-variable logic and for defining equations of polyadic equality algebras are given. The proof uses a family of nonrepresentable polyadic equality algebras ${\cal A}_n$ that are more and more nearly representable as $n$ increases: their $n$-generated subalgebras as well as their proper reducts are representable. The lattice of subvarieties of $RPEA_α$ is investigated and new open problems are asked about the interplay between the transposition operations and about generalizability of the results to infinite dimensions.

💡 Research Summary

The paper investigates the algebraic treatment of variable transposition in finite‑variable first‑order logic. Starting from the Lindenbaum‑Tarski formula algebra Fₘ of a finite‑variable logic, the authors enrich it with a unary operation p₍ᵢⱼ₎ that swaps the two variables vᵢ and vⱼ in any formula. The resulting structure Fₘ⁺ carries the usual Boolean operations, cylindrifications cᵢ (existential quantification), diagonal constants dᵢⱼ (equality), together with the transposition operations. Basic identities (P1)–(P7) describe how p₍ᵢⱼ₎ distributes over the Boolean connectives, interacts with cylindrifications and diagonals, and satisfies the group‑theoretic laws of the symmetric group on the index set. However, these identities alone do not capture the semantic equivalence ≡ on formulas; for instance, the equation (P8) holds in the quotient algebra Fₘ⁺/≡ but not in Fₘ⁺.

The problem, originally posed by James Johnson in 1969, asks whether the equational theory of Fₘ⁺/≡ can be axiomatized by a finite set of equations that uses the transposition operations only finitely many times, i.e., whether the transposition operations are finitely axiomatizable over the equational theory of the underlying cylindric algebra (the algebra of formulas without transpositions). Monk’s earlier result showed that even the cylindric part alone is not finitely axiomatizable; Johnson’s result extended this to the full polyadic equality algebras PSE_α (the algebraic counterpart of Fₘ⁺/≡). The present paper settles Johnson’s question in the negative by proving a much stronger statement.

Main theorem (Theorem 1). Every equational axiom system for the equations true in Fₘ⁺/≡ must, for each natural number n, contain an equation that involves at least n distinct algebraic variables together with at least one transposition, one cylindrification, and one diagonal constant. In other words, no finite or even “locally finite” axiom set can capture the full equational theory; the transposition operations force an unbounded number of variables and the simultaneous presence of the other two kinds of extra‑Boolean operations.

To establish this, the authors construct, for each odd prime power p ≥ 3, a polyadic‑type algebra A_p with the following properties (Theorem 3): 1. A_p is not isomorphic to any representable polyadic equality set algebra (hence non‑representable). 2. Every n‑generated subalgebra of A_p is isomorphic to a representable polyadic equality algebra. 3. The cylindrification‑free, diagonal‑free, and transposition‑free reducts of A_p are each isomorphic to subalgebras of representable algebras.

Thus A_p is “almost representable”: any finitely generated fragment looks representable, but the whole algebra does not. The non‑representability is witnessed by a specific equation e_p that holds in all representable algebras but fails in A_p; moreover, for q ≠ p, the equation e_q holds in A_p. Consequently, any axiom system that tries to capture all valid equations must contain infinitely many distinct equations, each involving more variables than any fixed bound, and each must involve the three extra‑Boolean operations. This yields Theorem 2, the concrete formulation of Theorem 1 for the class PSE_α (3 ≤ α < ω).

The paper also discusses several corollaries and extensions:

• A single transposition operation (e.g., P₀₁) already exhibits the same hardness; the class of algebras with only that operation cannot be axiomatized by a set of equations where P₀₁ appears only finitely many times.
• The lattice of subvarieties between the representable cylindric algebras Cs_α and the representable polyadic equality algebras PSE_α is shown to be rich; there are continuum many distinct equational theories lying strictly between them.
• Implications for proof systems of finite‑variable logic are drawn: any proof calculus that internalizes variable swapping must inherently use infinitely many schemata, reflecting the algebraic impossibility of a finite axiomatization.
• The authors outline open problems concerning the extension of these results to infinite dimensions (α ≥ ω) and to first‑order axiomatisations rather than equational ones.

In summary, the work demonstrates that the operation of swapping variables—though syntactically simple—introduces an essential source of algebraic complexity. The transposition operations cannot be tamed by a finite or even locally finite set of equations; they force an unbounded interaction with cylindrifications and diagonal constants. This resolves a long‑standing open problem in algebraic logic, clarifies the relationship between polyadic and cylindric algebras, and highlights deep connections to logical proof theory and database theory (where transpositions correspond to column renamings). The techniques—building “nearly representable” algebras whose finitely generated parts are representable—provide a powerful method that may be applicable to other algebraic structures exhibiting similar “local‑global” gaps.