Atomic Propositions

• Atomic propositions are minimal, irreducible units of information that serve as the foundational elements in formal logic, natural language processing, and quantum theory.
• They are rigorously defined through algebraic semiring formulations, grammatical clause decomposition, and closed subspace representations in Hilbert spaces.
• Their applications span deductive reasoning, knowledge extraction, and quantum contextuality proofs, offering both theoretical insights and practical inference tools.

An atomic proposition is a minimal, semantically autonomous unit of information within a logical, linguistic, or computational system. It is irreducible in the sense that it cannot be further decomposed without loss of information or introduction of unwarranted specificity. Atomic propositions serve as the foundational building blocks for more complex formulae and deduction procedures in formal logic, natural language understanding, and the semantics of quantum theory.

1. Formal Definitions Across Domains

Atomic propositions admit precise formal definitions, parameterized by the system in which they occur.

• Propositional Logic (Algebraic/Semiring Formulation): An atomic proposition (also: "atom") is a primitive element $x_i$ from a fixed countable set $X = \{ x_1, x_2, \ldots \}$, each with a formal complement $x_i^{\complement}$. No additional logical structure or axioms hold for atoms beyond idempotence and complementation rules imposed by the algebraic semiring with complement. All propositional formulae are constructed finitely from atoms and their complements using conjunction ($\cdot$) and disjunction ($\circ$) (Li et al., 2024).
• Semantic Information Theory and NLP: Atomic propositions are defined as minimal, semantically self-contained pieces of information extracted from natural language, such that they cannot be further subdivided without loss or hallucination of facts. Formally, in this context, a proposition $\varphi$ is atomic if and only if it is a single clause in conjunctive normal form (CNF), that is, $\varphi \equiv L_1 \vee \ldots \vee L_k$ where each $L_i$ is a literal (Pommeret et al., 3 Apr 2026). In natural language inference (NLI), atomic propositions are grammatical, strictly entailed statements obtained via decomposition of a hypothesis $H$ into constituent information units $a_1, \ldots, a_n$ with $X = \{ x_1, x_2, \ldots \}$0 (Srikanth et al., 12 Feb 2025).
• Quantum Logic: An atomic (elementary) proposition corresponds to a nontrivial closed linear subspace of a separable Hilbert space $X = \{ x_1, x_2, \ldots \}$1; equivalently, the range of a projection operator $X = \{ x_1, x_2, \ldots \}$2 on $X = \{ x_1, x_2, \ldots \}$3, with $X = \{ x_1, x_2, \ldots \}$4 (the set of all such subspaces or projections) (Bolotin, 2019). In Kochen-Specker-type constructions, the atomic proposition is identified with the occurrence of a specific outcome when commuting rank-1 projectors are measured, represented by the ray $X = \{ x_1, x_2, \ldots \}$5 in $X = \{ x_1, x_2, \ldots \}$6 (Cabello et al., 2018).

2. Construction and Generation Methodologies

Propositional Logic (Semirings)

The system is built using a free commutative semiring Rig$X = \{ x_1, x_2, \ldots \}$7, with the elements (atoms and their complements) combined under operations corresponding to logical conjunction and disjunction. Complement axioms—$X = \{ x_1, x_2, \ldots \}$8, $X = \{ x_1, x_2, \ldots \}$9, idempotence—are imposed to form the quotient semiring $x_i^{\complement}$0 (Li et al., 2024). Deduction is mechanized via Gröbner–Shirshov bases, covering all polynomials subject to these congruence relations.

Natural Language Processing and Knowledge Extraction

Atomic proposition extraction proceeds in several algorithmic stages:

1. Exemplar Construction: Manually constructed (sentence, atom-list) pairs illustrate correct decomposition into atoms.
2. Model-Based Generation: LLMs, e.g., Qwen3-32B or Llama-3-8b-instruct, are prompted with exemplars to generate candidate atoms for each hypothesis or sentence.
3. Pruning and Validation: Filters or check models retain only those candidates entailed by the source and (when required) not redundant with the premise.
4. Human Validation: Annotators validate grammaticality, informativeness, and entailment. Validity rates approach 95.7% with inter-annotator agreement $x_i^{\complement}$1 (Srikanth et al., 12 Feb 2025).
5. Recursive Atomization: For full decomposition, a propositioner model applies recursively until each output is atomic by the CNF criterion (Pommeret et al., 3 Apr 2026).

Quantum Propositions and Contextuality Constructions

Here, atomic propositions are instantiated as rays in Hilbert space and organized into contexts (mutually orthogonal sets). Graph-theoretic algorithms enumerate minimal true-implies-false (TIFS) and true-implies-true (TITS) sets, providing structurally minimal proofs of quantum contextuality. Nonisomorphic graphs subject to exclusiveness and completeness constraints are generated exhaustively up to the established minimal sizes (Cabello et al., 2018).

3. Taxonomy and Typology

Atomic propositions span varied semantic and mathematical classes:

Domain	Typical Form	Key Features
Propositional Logic	Primitive variable $x_i^{\complement}$2 or $x_i^{\complement}$3	Point-like, no internal structure
NLP/NLI	Minimal grammatical sentences or clauses	Entity/existence, event/action, attribute, spatial roles
Quantum Logic	Closed subspace (ray) $x_i^{\complement}$4	Orthogonality/exclusivity, context dependency

In NLP/NLI decomposition, typological categories include entity existence (“There is a man”), event assertions (“The person is riding”), property assertions (“The dog is brown”), thematic-role assignments (“Agent is the juggler”), and spatial relations (“They ride on a sidewalk”) (Srikanth et al., 12 Feb 2025). In quantum logic, classification is structural (based on orthogonality graphs and cliques) and linked to measurement settings and outcome exclusivity (Cabello et al., 2018).

4. Role in Inference, Deduction, and Proof

Atomic propositions are central to proof systems and explainability frameworks.

• Algebraic Deduction: In semiring-based propositional logic, deductive entailment $x_i^{\complement}$5 reduces to polynomial reduction: $x_i^{\complement}$6 is reducible to $x_i^{\complement}$7 modulo the Gröbner–Shirshov basis and relations $x_i^{\complement}$8 for $x_i^{\complement}$9 (Li et al., 2024).
• NLI and Knowledge Extraction: Decomposition into atoms permits the formulation of atomic inference subproblems (e.g., assess the status of $\cdot$0 or $\cdot$1 tuples), which enables targeted evaluation of an LLM’s logical consistency and inferential capacity (Srikanth et al., 12 Feb 2025).
• Quantum Contextuality: Minimal sets of atomic propositions are proven to be sufficient to refute noncontextual hidden-variable theories, with theorems showing, for example, that every minimal TIFS set in $\cdot$2 contains exactly $\cdot$3 atoms, and TITS $\cdot$4 atoms for $\cdot$5 (Cabello et al., 2018).

5. Truth Assignment and Semantics

Across logical paradigms, assignment of truth values to atomic propositions displays fundamental domain-dependence:

• Classical Logic: Each atom is assigned a classical truth value ($\cdot$6), underpinning Boolean semantics and the entire propositional calculus (Li et al., 2024).
• Quantum Logic: For $\cdot$7, no total dispersion-free (0/1-valued) valuation exists for all atomic propositions due to the algebraic structure of Hilbert space. Instead, atomic propositions bear probabilistic truth values, represented by probability measures $\cdot$8 such that $\cdot$9 in a given quantum state $\circ$0 (Bolotin, 2019). This codifies quantum indeterminacy and the impossibility of noncontextual hidden-variable models.
• NLP/Knowledge Extraction: Truth value is determined via entailment: an atomic proposition is retained only if strictly supported by the source hypothesis (for NLI) or by the sentence context (for information extraction), ensuring no hallucinated specificity (Pommeret et al., 3 Apr 2026).

6. Applications and Empirical Performance

Atomic propositions act as intermediaries in a variety of information processing pipelines and theoretical proofs.

• Knowledge Graph Construction: Integration of atomic proposition generation before triplet extraction improves recall in weak extractors (GLiREL, CoreNLP, Qwen3-0.6B), as measured on the SMiLER, FewRel, DocRED and CaRB benchmarks (Pommeret et al., 3 Apr 2026). For stronger LLMs, union or fallback strategies maintain or improve extraction performance.
• Auditability and Interpretability: Atomic propositions align with explainability aims; each carries no hidden conjunctions or referents, allowing users or models to directly trace which facts underpin system outputs (Pommeret et al., 3 Apr 2026).
• Quantum Foundations: Minimal TIFS/TITS sets derived from atomic propositions pinpoint the threshold for contextuality proofs and form the backbone for the combinatorial structure of Kochen–Specker and Hardy-type theorems (Cabello et al., 2018).
• Deductive Reasoning: Algebraic semiring methods based on atoms provide finite algorithms (subject to combinatorial scaling) for propositional deduction, framing both practical proof search and complexity lower bounds (Li et al., 2024).

7. Open Problems and Theoretical Significance

Despite the foundational status of atomic propositions, several domains face intrinsic limitations:

• Complexity: In propositional logic, the space of monomials and thus exhaustive deduction scales as $\circ$1 for $\circ$2 atoms, constraining practical scalability for large systems (Li et al., 2024).
• Quantum Nonclassicality: The necessity of probabilistic (rather than bivalent) semantics for atomic propositions in quantum logic not only precludes hidden variable models but also underscores the non-Boolean structure of physical reality at quantum scales (Bolotin, 2019).
• NLP Generalization: While atomic propositions yield gains in certain NLP pipelines, balancing the granularity of decomposition against entity coverage and preserving temporal, referential, and causal structure remain active research fronts (Srikanth et al., 12 Feb 2025, Pommeret et al., 3 Apr 2026).

A unifying implication is that atomic propositions, precisely defined for the domain at hand, serve as the irreducible foundation for inference, proof, and data-centric explainability, but their utility and semantics are tightly coupled to the algebraic and interpretive context in which they are embedded.