Logic of Partitions: Duality & Applications

• Logic of partitions is a framework that treats partitions as a dual to subset logic by using distinctions (dits) instead of membership to encode relationships.
• The framework defines algebraic operations like join, meet, implication, and negation, thus extending classical Boolean operations to partition structures.
• It underpins diverse applications from automata theory to quantum observables, revealing insights into both epistemic contexts and C*-algebraic decompositions.

Logic of partitions is the logical framework dual to Boolean subset logic, with partitions, quotient sets, and equivalence relations as interchangeable presentations, and with distinctions rather than membership as the basic semantic notion. In this framework, ordinary Boolean logic of subsets is the logic of elements or “Its”, while partition logic is the logic of distinctions or “Dits”; subsequent work develops implication and negation on partitions, studies partition logics obtained by pasting Boolean algebras, lifts partitions to direct-sum decompositions and $C^*$-algebraic subsystem structures, and analyzes special partition families such as Ruspini partitions in Gödel logic (Ellerman, 2020, Ellerman, 2014, Dvurecenskij et al., 2018).

1. Duality, distinctions, and the basic semantic universe

A partition $\pi=\{B,B',\dots\}$ of a set $U$ is a family of nonempty subsets that are mutually exclusive and jointly exhaustive. Equivalently, a partition determines an equivalence relation on $U$: two elements are equivalent iff they lie in the same block. Partition logic is interpreted on the lattice $\Pi(U)$ of partitions on a fixed universe $U$, ordered by refinement, and this order can be expressed by inclusion of ditsets: for partitions $\sigma,\pi$ on $U$,

$$\sigma\precsim\pi \quad \text{iff} \quad \operatorname{dit}(\sigma)\subseteq \operatorname{dit}(\pi).$$

The top element is the discrete partition

$\mathbf{1}=\bigl\{\{u\}\bigr\}_{u\in U},$

and the bottom element is the indiscrete partition

$\pi=\{B,B',\dots\}$0

When $\pi=\{B,B',\dots\}$1, $\pi=\{B,B',\dots\}$2 collapses to a two-element algebra isomorphic to the usual Boolean truth-values (Ellerman, 2020).

The core semantic notions are distinctions and indistinctions. A distinction of $\pi=\{B,B',\dots\}$3 is an ordered pair $\pi=\{B,B',\dots\}$4 of elements lying in different blocks, and its ditset is

$\pi=\{B,B',\dots\}$5

An indistinction is a pair in the same block, with

$\pi=\{B,B',\dots\}$6

This replaces the element-membership semantics of subset logic by a distinction/indistinction semantics on $\pi=\{B,B',\dots\}$7 (Ellerman, 2020, Ellerman, 2024).

The duality with subset logic is presented as category-theoretic. Subsets are subobjects, while partitions are quotient objects; equivalently, the dual notion to a subset is a partition. The elementary form of the duality is the contrast between elements and distinctions, or Its and Dits. Subset logic asks whether an element lies in a subset. Partition logic asks whether a pair of elements is distinguished by a partition. This duality is also reflected in the canonical maps: subset inclusion yields injections, while partition refinement yields surjections between quotient structures (Ellerman, 2014, Ellerman, 2024).

A quantitative extension of the same viewpoint appears as logical entropy. For a partition $\pi=\{B,B',\dots\}$8, logical entropy is the normalized counting of distinctions: $\pi=\{B,B',\dots\}$9 For equiprobable finite $U$0, this is

$U$1

This gives a direct measure of information-as-distinctions, in contrast with probability as normalized counting of elements (Ellerman, 2024).

2. Algebraic operations: join, meet, implication, negation, and Boolean transport

The lattice operations on partitions are not pointwise Boolean operations on arbitrary relations. Join behaves simply at the level of ditsets: $U$2 Meet is more subtle: $U$3 where

$U$4

is the largest ditset contained in $U$5, and $U$6 is the reflexive, symmetric, transitive closure of $U$7. The resulting closure/interior calculus is explicitly not a topological closure/interior in the usual sense (Ellerman, 2020).

A decisive advance was the definition of implication on partitions. For partitions $U$8, the implication $U$9 can be described blockwise: scan each block $U$0; if $U$1 is contained in some block of $U$2, then $U$3 is discretized into singletons, and otherwise $U$4 remains unchanged. Relation-theoretically,

$U$5

It satisfies the partition analogue of Boolean implication: $U$6 An equivalent graph-theoretic account labels the edges of the complete graph $U$7 by $U$8 and $U$9, retains exactly those edges where implication is false, and takes connected components as the blocks of $\Pi(U)$0 (Ellerman, 2020).

Negation is defined by implication into the bottom partition: $\Pi(U)$1 Absolute negation is almost trivial: if $\Pi(U)$2, then $\Pi(U)$3; if $\Pi(U)$4, then $\Pi(U)$5. The underlying reason is the Common-Dits Theorem: any two non-empty ditsets overlap. The important negation notion is therefore relative negation,

$\Pi(U)$6

For fixed $\Pi(U)$7, the $\Pi(U)$8-negated partitions form a Boolean algebra

$\Pi(U)$9

the Boolean core of the interval $U$0. Inside this Boolean core one recovers weak excluded middle,

$U$1

while ordinary excluded middle generally fails (Ellerman, 2020).

A general graph-theoretic and closure-theoretic method extends this beyond join, meet, and implication. Any $U$2-ary Boolean operation on truth-values can be transported to partitions by labeling edges of $U$3 with the $U$4-values determined by the input partitions, retaining exactly the edges where the Boolean function evaluates to $U$5, and then taking connected components. The equivalent closure-theoretic method applies the Boolean formula to ditsets and inditsets and then restores a genuine partition by reflexive-symmetric-transitive closure or, dually, interior. This construction yields partition analogues of all Boolean truth-functional operations, although compounds of these operations need not collapse to the original 16 binary Boolean functions (Ellerman, 2019).

3. Pasted Boolean algebras, prime structures, and automata

A partition logic in the finite structural sense is obtained by fixing a nonempty set $U$6 and a family $U$7 of partitions of $U$8. For each $U$9, one considers the Boolean algebra $\sigma,\pi$0 generated by the blocks of $\sigma,\pi$1. The pasting of the family $\sigma,\pi$2 is called a partition logic $\sigma,\pi$3. Each $\sigma,\pi$4 is locally classical, but their union need not be globally Boolean. This is the standard “locally classical, globally nonclassical” architecture of partition logic (Dvurecenskij et al., 2018).

The algebraic counterpart is given by Boolean atlases, quasi orthoalgebras, and orthoalgebras. The central structural theorem is exact: a quasi orthoalgebra $\sigma,\pi$5 is isomorphic to a partition logic iff $\sigma,\pi$6 is prime. Primeness is equivalent to having a separating set of two-valued probability measures, or equivalently enough prime ideals to separate points. This yields a precise criterion for classical realizability: partition logics are globally non-Boolean, yet they often admit a fully classical set-theoretic realization and separating dispersion-free states (Dvurecenskij et al., 2018).

This viewpoint is operationalized in automata theory. For a Moore or Mealy automaton, an experiment $\sigma,\pi$7 induces an equivalence relation on the state set $\sigma,\pi$8 by observational indistinguishability,

$\sigma,\pi$9

and hence a partition $U$0. The experimentally decidable propositions form the Boolean algebra generated by this partition. Pasting the Boolean algebras of all experiments yields the automaton propositional calculus. Conversely, to every partition logic there exists an automaton $U$1 such that the induced automaton propositional calculus is exactly that partition logic (Dvurecenskij et al., 2018).

Canonical finite examples include the firefly logic, the Wright triangle, and the $U$2 bowtie-type structures. The Wright triangle is an orthoalgebra and a partition logic, but not an orthomodular poset; the Fano plane is a non-example because it is not prime and has no separating set of two-valued states. This sharpens the boundary between partition logic and more general finite non-Boolean logics (Dvurecenskij et al., 2018).

4. Epistemic partitions, complementarity, and non-Booleanity outside quantum ontology

A separate but closely related line begins from classical state spaces and coarse-grained observables. If $U$3 is a classical state space and $U$4 is non-injective, then epistemic indistinguishability

$U$5

induces a partition $U$6 of $U$7. For a single partition, the resulting partition algebra $U$8 is a Boolean set algebra. The non-Booleanity appears when several different partitions, especially dynamically unstable, non-generating ones, are combined. Their local Boolean algebras may overlap nontrivially without embedding into one global Boolean algebra, yielding partition logics, partition test spaces, or orthomodular lattices formed by pasting Boolean blocks along overlaps (Atmanspacher et al., 2015).

The dynamical refinement

$U$9

determines whether a partition is generating. Compatibility is defined by

$$\sigma\precsim\pi \quad \text{iff} \quad \operatorname{dit}(\sigma)\subseteq \operatorname{dit}(\pi).$$0

incompatibility by

$$\sigma\precsim\pi \quad \text{iff} \quad \operatorname{dit}(\sigma)\subseteq \operatorname{dit}(\pi).$$1

and complementarity by

$$\sigma\precsim\pi \quad \text{iff} \quad \operatorname{dit}(\sigma)\subseteq \operatorname{dit}(\pi).$$2

The source of non-Booleanity is therefore epistemic, contextual, dynamical, and structural: each coarse-graining is Boolean locally, but the family of observational contexts is not (Atmanspacher et al., 2015).

This framework has been extended to social measurement. Partition logics are there described as non-Boolean event structures obtained by pasting Boolean algebras, built from a finite latent state space

$$\sigma\precsim\pi \quad \text{iff} \quad \operatorname{dit}(\sigma)\subseteq \operatorname{dit}(\pi).$$3

and observational contexts as partitions

$$\sigma\precsim\pi \quad \text{iff} \quad \operatorname{dit}(\sigma)\subseteq \operatorname{dit}(\pi).$$4

The paper constructs six explicit examples from personnel assessment, survey framing, clinical diagnosis, espionage coordination, legal pluralism, and organizational auditing, yielding instances of the $$\sigma\precsim\pi \quad \text{iff} \quad \operatorname{dit}(\sigma)\subseteq \operatorname{dit}(\pi).$$5 bowtie, triangle, pentagon, and automaton partition logics. The central thesis is precise: different modes of inquiry can be incompatible even though the underlying system remains fully value-definite; complementarity in this sense does not entail contextuality or ontic indeterminacy (Svozil, 28 Mar 2026).

Within this social partition-logic framework, each single context yields a Boolean event algebra, while the pasted structure is generally non-Boolean. Classical probabilities are generated by convex mixtures of dispersion-free states, with

$$\sigma\precsim\pi \quad \text{iff} \quad \operatorname{dit}(\sigma)\subseteq \operatorname{dit}(\pi).$$6

and the set of such probabilities is identified with the vertex packing polytope $$\sigma\precsim\pi \quad \text{iff} \quad \operatorname{dit}(\sigma)\subseteq \operatorname{dit}(\pi).$$7 of the exclusivity graph. When the same graph admits a faithful orthogonal representation, one also has Born-type probabilities

$$\sigma\precsim\pi \quad \text{iff} \quad \operatorname{dit}(\sigma)\subseteq \operatorname{dit}(\pi).$$8

forming the theta body $$\sigma\precsim\pi \quad \text{iff} \quad \operatorname{dit}(\sigma)\subseteq \operatorname{dit}(\pi).$$9. This separation of logical structure from probabilistic realization is explicit in the formalism (Svozil, 28 Mar 2026).

5. Vector-space and $\mathbf{1}=\bigl\{\{u\}\bigr\}_{u\in U},$0-algebraic generalizations

The dual progression from partitions to quantum theory replaces set partitions by direct-sum decompositions. A direct-sum decomposition (DSD) of a vector space $\mathbf{1}=\bigl\{\{u\}\bigr\}_{u\in U},$1 is a family of nonzero subspaces $\mathbf{1}=\bigl\{\{u\}\bigr\}_{u\in U},$2 such that

$\mathbf{1}=\bigl\{\{u\}\bigr\}_{u\in U},$3

The collection $\mathbf{1}=\bigl\{\{u\}\bigr\}_{u\in U},$4 is a partial partition algebra: the meet of two DSDs is always defined, but the join exists only under a compatibility condition. If

$\mathbf{1}=\bigl\{\{u\}\bigr\}_{u\in U},$5

their proto-join is the set of nonzero intersections $\mathbf{1}=\bigl\{\{u\}\bigr\}_{u\in U},$6, and $\mathbf{1}=\bigl\{\{u\}\bigr\}_{u\in U},$7 means that this proto-join spans $\mathbf{1}=\bigl\{\{u\}\bigr\}_{u\in U},$8. When compatible,

$\mathbf{1}=\bigl\{\{u\}\bigr\}_{u\in U},$9

while the refinement order is the exact vector-space analogue of partition refinement. The meet-semilattice $\pi=\{B,B',\dots\}$00 is atomistic, with atoms given by binary DSDs (Ellerman, 2016).

Maximal DSDs are decompositions into one-dimensional subspaces. For a maximal DSD $\pi=\{B,B',\dots\}$01, the interval

$\pi=\{B,B',\dots\}$02

is a full partition logic, and the implication operation inside this interval is inherited from set partition logic. This is the setting in which DSD logic expresses measurement by arbitrary self-adjoint operators: an observable is represented not merely by a projection subspace but by its eigenspace decomposition. In this sense, the quantum logic of direct-sum decompositions extends the Birkhoff–von Neumann logic of subspaces from propositions to observables (Ellerman, 2016).

The pedagogical model QM/Sets realizes this over $\pi=\{B,B',\dots\}$03, with

$\pi=\{B,B',\dots\}$04

A real-valued attribute $\pi=\{B,B',\dots\}$05 induces the DSD

$\pi=\{B,B',\dots\}$06

and the Born rule becomes ordinary finite conditional probability: $\pi=\{B,B',\dots\}$07 Density matrices in this setting can be read directly from the indit relation of a partition, so measurement becomes the creation of distinctions (Ellerman, 2014, Ellerman, 2016).

A more recent decompositional theory reformulates partitions directly at the $\pi=\{B,B',\dots\}$08-algebraic level. A global finite-dimensional quantum system is described by a $\pi=\{B,B',\dots\}$09-algebra $\pi=\{B,B',\dots\}$10, and a subsystem by a sub-$\pi=\{B,B',\dots\}$11-algebra $\pi=\{B,B',\dots\}$12. The distinctive claim is that one should allow non-factor subsystems, since symmetry and superselection naturally produce block-diagonal, non-factor algebras. This introduces Failure Of Local Tomography: the algebra generated by local subsystem algebras need not recover the whole global algebra (Vanrietvelde et al., 27 Jun 2025).

For a finite label set $\pi=\{B,B',\dots\}$13, a partition of $\pi=\{B,B',\dots\}$14 is a map

$\pi=\{B,B',\dots\}$15

with $\pi=\{B,B',\dots\}$16, $\pi=\{B,B',\dots\}$17, and, for disjoint $\pi=\{B,B',\dots\}$18, the condition that $\pi=\{B,B',\dots\}$19 in a blockwise bipartition sense. The centers $\pi=\{B,B',\dots\}$20 are central to the theory; in particular, the center of any conjunction is determined by the individual centers and the global center: $\pi=\{B,B',\dots\}$21 Every such partition has a routed-circuit Hilbert-space representation, but some are not fully representable: the fermionic local-mode partition is a valid algebraic partition for $\pi=\{B,B',\dots\}$22, yet necessarily retains residual pseudo-nonlocality in Hilbert-space representation (Vanrietvelde et al., 27 Jun 2025).

6. Topological, fuzzy, and generative extensions

In the Stone-dual setting of primitive Boolean spaces, partitions appear as trim $\pi=\{B,B',\dots\}$23-partitions, topological realizations of Hanf–Pierce structure diagrams. For a fixed primitive Stone space $\pi=\{B,B',\dots\}$24, the class of trim partitions is organized by regular refinement. If $\pi=\{B,B',\dots\}$25, then $\pi=\{B,B',\dots\}$26 is a surjective morphism of the underlying PO systems; conversely, a surjective morphism determines a unique coarser trim partition. The canonical trim partition $\pi=\{B,B',\dots\}$27 realizes the structure diagram $\pi=\{B,B',\dots\}$28, and the bounded trim partitions of $\pi=\{B,B',\dots\}$29 form a quasi-ordered system isomorphic to the corresponding class of extended PO systems. Rank partitions are the ideal completions of trim partitions, and for a primitive $\pi=\{B,B',\dots\}$30-Stone space the rank partitions are precisely the partitions of the form $\pi=\{B,B',\dots\}$31 for a trim partition $\pi=\{B,B',\dots\}$32 (Apps, 2023).

A distinct line studies fuzzy partitions in Gödel logic. A Ruspini partition is a finite family of fuzzy sets $\pi=\{B,B',\dots\}$33, $\pi=\{B,B',\dots\}$34, such that

$\pi=\{B,B',\dots\}$35

The paper shows that this additive condition is not directly expressible in Gödel logic $\pi=\{B,B',\dots\}$36, because the language has no connective for arithmetic addition. What is exactly captured is the weaker order-invariant notion of weak Ruspini partition, characterized by comparison maps, order isomorphisms, and a finite forest of order types. The theory also isolates the $\pi=\{B,B',\dots\}$37-overlapping condition

$\pi=\{B,B',\dots\}$38

for distinct $\pi=\{B,B',\dots\}$39, proves that it is expressible already in four-valued Gödel logic $\pi=\{B,B',\dots\}$40, and gives constructive normal forms for the canonical Gödel formula associated with a partition (Codara et al., 2014).

More recently, finite partition logics with a separating set of two-valued states have been translated into Simple Generative Logic Grammars implemented in Prolog. If $\pi=\{B,B',\dots\}$41 has atoms $\pi=\{B,B',\dots\}$42 and separating two-valued states $\pi=\{B,B',\dots\}$43, with supports

$\pi=\{B,B',\dots\}$44

then the grammar is defined by

$\pi=\{B,B',\dots\}$45

The flagship example is the five-atom V-logic $\pi=\{B,B',\dots\}$46, with contexts

$\pi=\{B,B',\dots\}$47

whose partition-logical incidence structure is rendered as the “Quantum Square.” The paper is explicit that the logical structure and the perceptible realization remain distinct; this makes partition logic available as a generative score while preserving its atom-state incidence structure (Jendreiko et al., 19 Mar 2026).

Taken together, these developments suggest that logic of partitions is not a single formalism but a family of dual, order-theoretic, algebraic, topological, fuzzy, and quantum constructions organized around one persistent idea: structure can be encoded by distinctions, by the compatibility or incompatibility of contexts, and by the ways a whole can be partitioned without reducing that structure to ordinary Boolean membership (Ellerman, 2014, Vanrietvelde et al., 27 Jun 2025).