Nominal Alpha-Equivalence

• Nominal alpha-equivalence is a framework that defines equality for syntax with binders by identifying terms differing only by admissible renaming of bound names.
• It leverages nominal sets, where support and freshness arise from finite permutation actions on atoms, ensuring that semantic content is invariant under renaming.
• The methodology underpins mechanized proof principles, unification, substitution, rewriting, and coalgebraic semantics for languages with complex binding structures.

Nominal alpha-equivalence is the nominal-set and nominal-logic account of equality for syntax with binders modulo consistent renaming of bound names. In this setting, names are treated as atoms acted on by finite permutations, and binding is represented through abstraction objects whose equality identifies terms that differ only by admissible renaming of bound names. This yields a uniform framework for support, freshness, equivariance, abstraction, substitution, unification, rewriting, and semantics for languages with binding, while preserving a first-order presentation of syntax with names [0609062].

1. Nominal setting and the role of names

Nominal treatments begin with an infinite set of names or atoms and the group of finite permutations generated by transpositions. A nominal set is then a set equipped with a permutation action such that every element has finite support. In the standard formulation, a finite set $S$ supports $x$ when every permutation fixing $S$ pointwise also fixes $x$, and the least such support is written $\operatorname{supp}(x)$. Freshness is defined from support by $a \# x$ iff $a \notin \operatorname{supp}(x)$ (Paranhos et al., 30 Sep 2025, Milius et al., 2016, Kurz et al., 2013).

A central nominal observation is that support and freshness are not merely syntactic annotations but are induced by the permutation action itself. One consequence, made explicit in constructive work on nominal sets, is that support depends on the equality used on the underlying type: for $\lambda$-terms, under syntactic equality support is all occurring variables, whereas under $\alpha$-equivalence support is the set of free variables (Paranhos et al., 30 Sep 2025). This makes nominal alpha-equivalence an intrinsic notion rather than a derived side condition over raw syntax.

Equivariance is the companion principle to support. A relation or map is equivariant when it commutes with permutation action. In nominal developments this is the formal expression of the idea that semantic content does not depend on accidental choices of names. Freshness, abstraction, alpha-equivalence, coalgebraic structure maps, and many proof principles are explicitly proved equivariant in mechanized and coalgebraic treatments (Paranhos et al., 30 Sep 2025, 0809.3960, Frank et al., 2022).

2. Abstraction, freshness, and the nominal definition of alpha-equivalence

The basic nominal constructor for binding is abstraction. For a nominal set $X$, one forms the abstraction set $x$0, written as the quotient of $x$1 by an equivariant relation identifying pairs that differ only by renaming the bound atom. The abstraction class of $x$2 is written $x$3 or, in some mechanizations, $x$4. Its support satisfies

$x$5

capturing the fact that the abstracted name ceases to contribute to support (Paranhos et al., 30 Sep 2025, Frank et al., 2022, Milius et al., 2016).

A standard characterization of abstraction equality is

$x$6

which makes the renaming principle explicit: two abstractions are equal when one body becomes the other after swapping binders, provided the freshness side condition holds (Frank et al., 2022, 0809.3960).

Constructive nominal-set work states nominal alpha-equivalence on $x$7 in existential and universal fresh-name forms. The existential form is

$x$8

and an equivalent universal form quantifies over all sufficiently fresh $x$9 (Paranhos et al., 30 Sep 2025). This definition avoids explicit capture-avoiding renaming functions and instead reduces alpha-equivalence to common fresh swapping. The same pattern appears in nominal coalgebraic semantics for bar words, where $S$0-equivalence of strings with binding letters is generated from equality of abstraction objects (Frank et al., 2022).

Freshness itself admits a constructive reformulation in terms of swapping. One definition used in Rocq is

$S$1

together with an equivalent universal version quantifying over all $S$2 outside a support (Paranhos et al., 30 Sep 2025). This “some fresh / any fresh” equivalence is a recurring proof principle in nominal reasoning.

3. Raw named syntax, suspensions, and general binders

A characteristic feature of nominal syntax is that terms remain raw first-order terms with explicit names rather than quotient classes. In nominal unification, terms include atoms, abstractions, tuples, function-symbol applications, and suspensions $S$3, where a variable carries a delayed permutation (Urban, 2010, Ayala-Rincón et al., 2017). Equality modulo alpha-equivalence is then an inductively defined relation on these raw terms. For abstractions, the defining rule is the familiar nominal clause: $S$4 when $S$5, $S$6, and $S$7 (Urban, 2010). For suspensions, equality depends on the disagreement set of the permutations and the freshness environment (Urban, 2010, Ayala-Rincón et al., 2017).

This raw named presentation has two technical consequences. First, terms are not implicitly quotient types; alpha-equivalence is a relation over ordinary syntax (Urban, 2010). Second, substitutions may be capturing in the usual syntactic sense, because safety is recovered through freshness constraints and alpha-equivalence rather than through generating fresh names during substitution (Urban, 2010).

Nominal Isabelle extends this picture from single binders to general binders. It distinguishes three alpha-equivalence schemes for abstractions with multiple bound atoms: list binders, set binders, and set$S$8 binders. List binders preserve order; set binders ignore order but not vacuous binders; set$S$9 binders ignore both order and vacuous binders (Urban et al., 2012). The corresponding abstraction types satisfy support equations of the form

$x$0

or the set-of-list analogue, validating that the designated atoms are genuinely bound (Urban et al., 2012). This general-binder infrastructure is motivated by type schemes, let-groups, pattern binders, and Core Haskell/System $x$1 alternatives (Urban et al., 2012).

A useful contrast is provided by work on a direct inductive definition of $x$2-equivalence for named untyped $x$3-terms. That account uses paired binder contexts and shadow-sensitive lookup, but explicitly does not use nominal permutations, abstraction types, support, or freshness as primitive notions (Apinis et al., 14 Jul 2025). It addresses the same conceptual problem but is not nominal in the technical sense.

4. Mechanization and nominal proof principles

Mechanized nominal developments emphasize proof principles that internalize informal fresh-name reasoning. In Isabelle/HOL, the nominal datatype package represents syntax and even residuals of the $x$4-calculus up to alpha-equivalence from the outset. Binder equality is given by abstraction equality, for example

$x$5

and derived induction and case rules allow bound names to be chosen fresh for an arbitrary finite context (0809.3960). This is how the formalization recovers the usual variable convention in proofs of bisimulation and congruence.

Constructive Rocq formalizations pursue the same goal with different infrastructure. A major result is a reusable nominal hierarchy with type classes, setoids, generalized rewriting, support, freshness, abstraction, and alpha-equivalence, together with finitely supported functions and abstraction-based proof principles (Paranhos et al., 30 Sep 2025). Two of the most important are the Freshness Theorem and the Freshness Condition for Binders. The latter yields a unique function on abstraction classes from a function on representatives once an appropriate freshness condition is established, thereby replacing ad hoc representative choices with a nominal universal property (Paranhos et al., 30 Sep 2025).

Dependently sorted nominal signatures push the same ideas into a first-order dependent setting. There, sorts and terms may contain nominal abstractions, generalized concretion acts as an elimination form, substitution is capture-avoiding atom substitution, and the sorting discipline is explicitly closed under alpha-conversion (Fernández et al., 14 Oct 2025). The framework proves equivariance of sorting, freshness stability under alpha, congruence of alpha-equivalence, and compatibility of substitution with alpha-equivalence, with the stated aim of serving as a basis for a future type theory in which raw syntax modulo alpha-equivalence is handled natively (Fernández et al., 14 Oct 2025).

5. Coalgebraic, automata-theoretic, and infinitary semantics

Nominal alpha-equivalence is also a semantic organizing principle. In nominal coalgebraic data types, raw functors with binding positions $x$6 are replaced by quotient functors using abstraction $x$7. For $x$8-syntax this yields

$x$9

whose initial algebra gives finite $\operatorname{supp}(x)$0-terms modulo alpha-equivalence and whose final coalgebra captures infinitary terms with finitely many free variables modulo alpha-equivalence (Kurz et al., 2013, Milius et al., 2016). For mixed inductive-coinductive signatures, the same nominal abstraction construction yields terminal coalgebras of alpha-equivalence classes of infinitary terms with finitely many free variables, together with capture-avoiding substitution defined directly on the classes (Cerda, 4 Nov 2025).

The rational or regular fragment is described coalgebraically by rational fixpoints. For binding functors such as $\operatorname{supp}(x)$1, the rational fixpoint consists of those alpha-equivalence classes of $\operatorname{supp}(x)$2-trees that contain at least one rational raw tree (Milius et al., 2016). This is the nominal counterpart of regular behavior with binding.

Automata models with binders also rely on nominal alpha-equivalence. Regular nondeterministic nominal automata accept bar languages, that is, sets of bar words modulo alpha-equivalence, and their bound transitions are required to be alpha-invariant: $\operatorname{supp}(x)$3 The accepted semantics of a state is therefore a set of equivalence classes $\operatorname{supp}(x)$4, not a set of chosen representatives (Frank et al., 2022). A related line of work distinguishes two readings of the same alpha-quotiented syntax for data languages: restricting to clean representatives yields a global-freshness semantics, while allowing unrestricted representatives yields local freshness as a quotient of the global semantics (Schröder et al., 2016).

6. Unification, fixed points, rewriting, and equational theories

Nominal unification computes substitutions making raw named terms equal modulo nominal alpha-equivalence (Urban, 2010). Its distinctive features are that it uses raw first-order terms with names, suspensions, freshness environments, and swappings, and that it does not require generating fresh names during unification (Urban, 2010). When abstractions with different binders are unified, the problem is reduced to unifying the body against a swapped body together with a freshness side condition, directly reflecting the nominal definition of alpha-equivalence (Urban, 2010, Ayala-Rincón et al., 2017).

A later development replaces freshness as the primitive notion by permutation fixed-point constraints. The key equivalence is that freshness can be derived from fixed points: $\operatorname{supp}(x)$5 and alpha-equivalence can be axiomatized using fixed-point side conditions instead of freshness side conditions (Ayala-Rincón et al., 2019). This formulation is equivalent to standard nominal alpha-equivalence on ordinary syntax, but it behaves better in the presence of equational theories such as commutativity, because a permutation may fix a term modulo $\operatorname{supp}(x)$6 without the corresponding atoms being fresh (Ayala-Rincón et al., 2019).

That observation leads to nominal $\operatorname{supp}(x)$7-unification and to nominal rewriting and narrowing modulo equational theories. Nominal $\operatorname{supp}(x)$8-unification extends standard nominal unification to commutative symbols and reduces problems to families of fixed-point equations solvable by permutation combinatorics (Ayala-Rincón et al., 2017). Nominal equational rewriting and narrowing then define rewriting modulo both alpha-equivalence and an equational theory $\operatorname{supp}(x)$9, prove nominal $a \# x$0-coherence as the property relating class-based and operational rewriting, and establish a nominal $a \# x$1-Lifting theorem connecting rewriting and narrowing sequences (Ayala-Rincón et al., 6 Jun 2025). The commutative theory $a \# x$2 is the concrete worked theory for which these results are fully instantiated (Ayala-Rincón et al., 6 Jun 2025).

7. Contrasts, extensions, and adjacent notions

Not every binder-sensitive equality on named syntax is nominal alpha-equivalence in the technical sense. The direct inductive context-indexed account of $a \# x$3-equivalence for named $a \# x$4-terms uses neither permutations nor support nor abstraction types (Apinis et al., 14 Jul 2025). Likewise, a calculus that explicitly separates variables from names may preserve ordinary alpha-equivalence for variables while denying any analogous local alpha-equivalence principle for names, because names are semantically relevant interface labels for nominal rebinding (Ancona et al., 2013). These approaches illuminate neighboring design choices rather than extending nominal alpha-equivalence itself.

A second adjacent development is context-sensitive $a \# x$5-equivalence for open subterms in $a \# x$6-terms. There the main issue is that ordinary alpha-equivalence treats free variables too primitively for subterm comparison in ambient context. The proposed equivalence is defined on term nodes $a \# x$7, characterized both as fork equivalence and as bisimilarity, and supports $a \# x$8 hashing of $a \# x$9-terms modulo that equivalence (Blaauwbroek et al., 2024). This is binder-aware and exact, but it is not a nominal-sets account.

A broader but weaker analogy appears in machine learning work on “interchangeable tokens,” where consistent renaming of semantically symmetric symbols is treated as a covariance problem and measured by an “alpha-covariance” metric. That work invokes alpha-equivalence as motivation but does not model binding scopes, free/bound distinction, or capture-avoiding renaming in the nominal sense (Işık et al., 2024). A plausible implication is that nominal alpha-equivalence continues to mark the boundary between purely renaming-robust treatments of symbols and full formal treatments of binding.

Across these contrasts, the technical core of nominal alpha-equivalence remains stable: finite permutations of atoms, support and freshness induced by permutation action, abstraction objects $a \notin \operatorname{supp}(x)$0, equality by common fresh swapping, and proof principles that make bound-name irrelevance mathematically explicit (Paranhos et al., 30 Sep 2025, Milius et al., 2016, Urban, 2010).