Logical Coproduct: Combining Deductive Systems

• Logical coproduct is a construction that integrates coproducts as a logical operation to combine deductive systems while preserving individual axioms and rules.
• It leverages internal logic from subobject structures in higher categories and algebraic representations via monads, quantales, and nuclei.
• This approach unifies categorical semantics with proof theory by modeling disjunction, additive operations, and the combination of logical theories.

Logical coproduct is the use or construction of a coproduct as a logical operation rather than merely as a formal colimit. In the cited literature, coproducts play this role in several tightly connected settings: as internal disjunction reconstructed from subobject structure in higher categories (Frey et al., 2021), as the categorical sum of monads or algebraic theories used to combine effects while identifying only the shared units (Adámek et al., 2014), and as the coproduct of deductive systems obtained by combining languages, axioms, and rules in a universal way (Russo, 26 Aug 2025). In proof theory, ordinary coproducts interpret additive disjunction, but attempts to read all co-intuitionistic or subtraction-like connectives through set-theoretic disjoint union fail, forcing a move to richer categorical semantics (Bellin, 2014).

1. Conceptual scope

A logical coproduct is a coproduct whose defining universal property is read through logical data. In one direction, it is the operation that combines theories or deductive systems without adding identifications beyond those already present. In another, it is the categorical realization of disjunction: the object that classifies a choice between alternatives and whose slices decompose accordingly. In algebraic semantics, the same idea appears as the sum of two theories, where variables remain the shared trivial terms and all nontrivial operations are combined freely (Adámek et al., 2014).

This use of coproduct is especially explicit in the order-theoretic treatment of deductive systems. A deductive system $S=(D,\vdash)$ over a propositional language $L$ is represented by a nucleus on the $P\Sigma_L$-module $PD$, with

$$\gamma_\vdash(\Phi)=\{\Psi\in PD\mid \Phi\vdash \Psi\},$$

and the fixed points $PD_\gamma$ are the lattice of theories. For systems over languages $L_1$ and $L_2$, the logical coproduct is formed on the disjoint union language $L=L_1\sqcup L_2$ by taking all axioms and inference rules from both systems; equivalently, if $\gamma_1,\gamma_2$ are the corresponding nuclei, the combined system is governed by

$L$0

This is the paper’s logical coproduct of deductive systems (Russo, 2021).

A closely related reading appears in category theory proper. In a locally Cartesian closed $L$1-category with a subobject classifier, coproducts are obtained from the internal logic of subobjects, finite joins, and pullback-stable reasoning. Here the coproduct is “logical” because it is reconstructed from the operations corresponding to $L$2, quantification, and truth values rather than postulated independently (Frey et al., 2021).

2. Internal disjunction in locally Cartesian closed $L$3-categories

A decisive higher-categorical result states that every locally Cartesian closed $L$4-category with subobject classifier has a strict initial object and disjoint, universal binary coproducts (Frey et al., 2021). Equivalently, for any finite family $L$5, the coproduct exists and pullback along the coproduct inclusions induces an equivalence

$L$6

This is the higher-categorical form of extensivity.

The construction proceeds through subobject lattices. For each object $L$7, the subobject poset $L$8 is a Heyting semilattice; pullback preserves finite meets and Heyting implication; and the right adjoint to pullback gives a form of universal quantification. With a subobject classifier, finite joins exist in every $L$9. The initial object is recovered from the least subobject of the terminal object, and binary coproducts are then built by embedding two objects as disjoint subobjects inside a larger ambient object and taking their join.

The result is called logical because the entire coproduct structure is extracted from internal logic. Subobjects correspond to propositions, meets to conjunction, joins to disjunction, implication to exponentials in slices, quantification to adjoints to pullback, and the classifier $P\Sigma_L$0 to truth values. On this view, the coproduct $P\Sigma_L$1 is a logical disjunction of subobjects, and the slice equivalence

$P\Sigma_L$2

expresses the universal and disjoint character expected of a logical sum (Frey et al., 2021).

3. Coproducts of deductive systems and logical combination

The ordered-algebraic treatment of logical combination uses quantales, modules, and nuclei. For a propositional language $P\Sigma_L$3, the substitution monoid $P\Sigma_L$4 gives rise to the quantale $P\Sigma_L$5, which acts on powersets of formulas, equations, or sequents. In this setting, consequence relations are exactly $P\Sigma_L$6-module nuclei on $P\Sigma_L$7, establishing the correspondence

$P\Sigma_L$8

This identifies syntax, substitution, and deductive closure within a single categorical structure (Russo, 2021).

Given deductive systems $P\Sigma_L$9 and $PD$0 of the same syntactic type, the logical coproduct is defined on the disjoint union language $PD$1. Its consequence relation is generated by all axioms and rules of the two systems. The corresponding theory modules of the expanded logics satisfy

$PD$2

and the coproduct of the original theory modules embeds as a sup-lattice,

$PD$3

This shows that the logical combination contains faithful copies of the original deductive structures while remaining governed by a universal construction (Russo, 2021).

A later categorical refinement packages this construction into the category $PD$4 of propositional deductive systems. Its objects are pairs $PD$5, and its morphisms are those quantale-module morphisms whose quantale part is induced by a language translation. In this category, the logical coproduct is not merely analogous to a coproduct: it is the coproduct. If $PD$6 is the disjoint union of the component languages and $PD$7 is the theory module of the deductive system generated by the union of all component axioms and rules, then

$PD$8

is, up to isomorphism, the coproduct of the family $PD$9 in $$\gamma_\vdash(\Phi)=\{\Psi\in PD\mid \Phi\vdash \Psi\},$$0 (Russo, 26 Aug 2025).

This resolves an important ambiguity. The quantale-module coproduct in the ambient category $$\gamma_\vdash(\Phi)=\{\Psi\in PD\mid \Phi\vdash \Psi\},$$1 is not, in general, the same as the logical coproduct relevant to deductive systems. The latter becomes genuinely categorical only after restricting to the morphisms that represent interpretations and language translations (Russo, 26 Aug 2025).

4. Sums of theories, bialgebras, and monads on $$\gamma_\vdash(\Phi)=\{\Psi\in PD\mid \Phi\vdash \Psi\},$$2

For monads on $$\gamma_\vdash(\Phi)=\{\Psi\in PD\mid \Phi\vdash \Psi\},$$3, coproducts formalize “combining effects” and, under the algebraic-theory reading, forming the sum of two theories while identifying only the shared trivial terms or units (Adámek et al., 2014). If $$\gamma_\vdash(\Phi)=\{\Psi\in PD\mid \Phi\vdash \Psi\},$$4 and $$\gamma_\vdash(\Phi)=\{\Psi\in PD\mid \Phi\vdash \Psi\},$$5 are monads, their coproduct $$\gamma_\vdash(\Phi)=\{\Psi\in PD\mid \Phi\vdash \Psi\},$$6 is the monad of free $$\gamma_\vdash(\Phi)=\{\Psi\in PD\mid \Phi\vdash \Psi\},$$7-bialgebras. Kelly’s observation is that such a coproduct exists exactly when every set generates a free bialgebra.

The decisive construction in the consistent case is expressed through unit complements. A monad $$\gamma_\vdash(\Phi)=\{\Psi\in PD\mid \Phi\vdash \Psi\},$$8 is consistent when every unit map $$\gamma_\vdash(\Phi)=\{\Psi\in PD\mid \Phi\vdash \Psi\},$$9 is injective. For such a monad one defines

$PD_\gamma$0

the nontrivial part of $PD_\gamma$1, regarded as a functor on $PD_\gamma$2. For two consistent monads $PD_\gamma$3 and $PD_\gamma$4, one studies the recursive system

$PD_\gamma$5

If for every set $PD_\gamma$6 this system has an initial algebra $PD_\gamma$7 in $PD_\gamma$8, then the coproduct exists and is given by

$PD_\gamma$9

The converse also holds: if $L_1$0 exists, then these initial algebras must exist (Adámek et al., 2014).

The logical interpretation is direct. For a theory $L_1$1, the associated monad $L_1$2 sends a set of variables to the set of terms modulo equations. Then $L_1$3 is the monad of the combined theory $L_1$4, where the common units correspond to variables. The formula

$L_1$5

describes terms as alternating layers of the nontrivial parts of the two theories, with the final $L_1$6 separating off the variable-only terms (Adámek et al., 2014).

This algebraic reading also yields a conservativity statement. If $L_1$7 and $L_1$8 are consistent and $L_1$9 exists, then the coproduct embeddings are injective. More generally, a coproduct of injective monad morphisms is injective. In theory language, this means that adding the second theory does not collapse distinct elements coming from the first. Existence is sharply controlled by fixpoints: two consistent monads have a coproduct iff either one is substantially exceptional or they have arbitrarily large common fixpoints $L_2$0 with

$L_2$1

A consistent monad has a coproduct with every monad iff it is substantially exceptional (Adámek et al., 2014).

5. Proof theory, additive disjunction, and semantic limitations

In categorical proof theory, coproducts interpret additive disjunction. For the extension $L_2$2, a category $L_2$3 with coproducts supports the familiar term formers

$L_2$4

and thus gives semantics for additive disjunction in the standard categorical sense (Bellin, 2014).

A common misconception is that this exhausts the disjunctive role of coproducts. The co-intuitionistic literature shows that it does not. The co-exponent of $L_2$5 and $L_2$6 is defined as an object $L_2$7 equipped with

$L_2$8

such that every $L_2$9 factors uniquely through $L=L_1\sqcup L_2$0. In $L=L_1\sqcup L_2$1, however, this co-exponent exists if and only if $L=L_1\sqcup L_2$2 or $L=L_1\sqcup L_2$3. The obstruction is that coproducts in $L=L_1\sqcup L_2$4 are disjoint unions, so morphisms into $L=L_1\sqcup L_2$5 make a pointwise choice of summand, and the universal factorization demanded by the co-exponent cannot be satisfied nontrivially (Bellin, 2014).

The categorical remedy is to move from cartesian structure to symmetric monoidal left-closed categories, where subtraction is modeled by a left adjoint

$L=L_1\sqcup L_2$6

In this setting, the relevant “disjunctive” connective is not set-theoretic disjoint union but the monoidal structure associated with co-intuitionistic linear logic (Bellin, 2014).

A distinct proof-theoretic result shows how far coproduct logic can simulate product logic. Let $L=L_1\sqcup L_2$7 be the category with finite, possibly empty, coproducts freely generated by one object, and $L=L_1\sqcup L_2$8 the category with finite, possibly empty, products freely generated by a countable set of objects. A skeleton of $L=L_1\sqcup L_2$9 has a subcategory isomorphic to a skeleton of $\gamma_1,\gamma_2$0, and therefore $\gamma_1,\gamma_2$1 has a subcategory equivalent to $\gamma_1,\gamma_2$2. Proof-theoretically, deductions of pure conjunctive logic with a countable set of propositional letters can thus be represented by deductions in pure disjunctive logic with one propositional letter. By taking opposite categories, the dual result exchanges products and coproducts, conjunction and disjunction (Dosen et al., 2015).

6. Structural consequences and related categorical phenomena

Once coproducts are available, they interact strongly with other logical structure. In a (weakly) Cartesian closed category with finite coproducts, there is a distributivity isomorphism

$\gamma_1,\gamma_2$3

One direction is the canonical map

$\gamma_1,\gamma_2$4

and the opposite direction is obtained by exploiting the adjunction $\gamma_1,\gamma_2$5. The result expresses that product with $\gamma_1,\gamma_2$6 preserves binary coproducts, making categorical disjunction distribute over conjunction in the expected manner (Benini, 2014).

In monoidal settings, coproducts may strengthen further into biproducts. If $\gamma_1,\gamma_2$7 has an initial object $\gamma_1,\gamma_2$8, the coproduct $\gamma_1,\gamma_2$9, and each endofunctor $L$00 preserves both $L$01 and $L$02, then $L$03 is semi-additive iff $L$04 and $L$05 admit right duals. When each $L$06 preserves finite coproducts, this yields the criterion that $L$07 has a zero object and finite biproducts iff the initial object $L$08 and the coproduct $L$09 admit right duals (Garner et al., 2015).

These results place logical coproducts within a wider hierarchy. At one end, coproducts behave as disjunctions reconstructed from subobjects and internal logic; at another, they are the universal sums of theories, monads, or deductive systems; and under stronger preservation and duality hypotheses they participate in distributive or even additive structures. At the same time, the co-intuitionistic examples show that logical coproduct should not be identified uncritically with disjoint union. Its meaning depends on the surrounding semantics: internal logic, substitution structure, algebraic syntax, or monoidal closure (Frey et al., 2021).