Semantic Element (SE) in Logic Programming

• Semantic Element (SE) is a framework defined via pairs of interpretations that provide a monotonic basis for non-monotonic reasoning in logic programming.
• It is used to determine strong program equivalence by transforming rules into a unique canonical form based on their SE-models.
• SE-models facilitate theoretical analysis of program updates, rule expressivity, and computational complexity in answer-set programming.

A semantic element (SE), in the context of non-monotonic logic programming and answer-set programming (ASP), refers to the strong equivalence model, also known as the SE-model or Here-and-There model. SE-models provide a semantic, monotonic framework to characterize logic programs, capturing both their classical models and their answer sets, and determining when two programs are strongly equivalent (interchangeable in all contexts) (Slota et al., 2013, Slota et al., 2011).

1. Formal Definition and Construction of SE-Models

Let $A$ be a finite set of propositional atoms. An SE-interpretation is a pair $X = (I,J)$ with $I \subseteq J \subseteq A$. For a (possibly disjunctive) logic program $\Pi$, one defines its Gelfond–Lifschitz reduct relative to $J$ by

$$\Pi^J = \{ H(r)^+ \leftarrow B(r)^+ \mid r \in \Pi, B(r)^- \subseteq J, B(r)^- \cap J = \emptyset \},$$

where $H(r) = (H(r)^+,H(r)^-)$ and $B(r) = (B(r)^+,B(r)^-)$ are the positive and negative literals in the head and body of each rule $r$. An SE-model of $\Pi$ is an SE-interpretation $X = (I,J)$0 such that $X = (I,J)$1 and $X = (I,J)$2. The set $X = (I,J)$3 denotes all SE-models of $X = (I,J)$4.

These models yield the following properties:

• $X = (I,J)$5 is monotonic in $X = (I,J)$6.
• $X = (I,J)$7 is an answer set of $X = (I,J)$8 iff $X = (I,J)$9 and there is no $I \subseteq J \subseteq A$0 with $I \subseteq J \subseteq A$1.
• $I \subseteq J \subseteq A$2 iff $I \subseteq J \subseteq A$3 and $I \subseteq J \subseteq A$4 are strongly equivalent.

At the rule level, a rule $I \subseteq J \subseteq A$5 over propositional alphabet $I \subseteq J \subseteq A$6 has the form $I \subseteq J \subseteq A$7. The SE-models of a rule are given via the above reduct-based satisfaction, providing canonical forms and fine control over equivalence classes (Slota et al., 2011).

2. SE-Models and Program Equivalence

Strong equivalence of logic programs is captured via SE-models. Programs $I \subseteq J \subseteq A$8 and $I \subseteq J \subseteq A$9 are strongly equivalent iff $\Pi$0. At the rule level, two rules $\Pi$1 and $\Pi$2 are SE-equivalent if they have the same set of SE-models.

A key result is the existence of a unique canonical rule for each SE-equivalence class: every rule $\Pi$3 is SE-equivalent to exactly one canonical rule $\Pi$4, where canonical rules have disjoint positive/negative head/body sets with required constraints:

• $\Pi$5 is canonical if it is the tautology $\Pi$6 or of the form $\Pi$7, with $\Pi$8 pairwise disjoint, and $\Pi$9 if $J$0.

The transformation to canonical form involves SE-preserving eliminations and reorganization, generating a unique representation for the SE-class (Slota et al., 2011).

3. Expressivity and Rule Representability

Not every set of SE-interpretations is representable as the set of SE-models of a single rule. The characterization theorem states that $J$1 is rule-representable if and only if its complement can be decomposed into unions of two convex sublattices in $J$2:

$J$3

This lattice-theoretic condition precisely picks out the expressible sets and allows for constructing rules yielding required SE-models (Slota et al., 2011).

4. SE-Models in Program Update Frameworks

The Katsuno–Mendelzon (KM) postulates for belief update (BU1–BU8) can be restated for SE-models, supported by the monotonic and closure properties of SE-model sets. Program conjunction and disjunction are defined so that $J$4 and $J$5, and "basic programs" (analogue to complete formulas) are those with exactly two SE-models $J$6.

Eight postulates (P1–P8) for rule-update operators on SE-models directly lift the KM update axioms. The representation theorem then states that any operator satisfying these postulates corresponds to a family of (semi-)faithful and organized preorder assignments $J$7 on SE-interpretations, such that

$J$8

A concrete example is the "Winslett-style" update, where the preference on SE-models minimizes change first in the second component ($J$9) and then, if tied, in the first ($$\Pi^J = \{ H(r)^+ \leftarrow B(r)^+ \mid r \in \Pi, B(r)^- \subseteq J, B(r)^- \cap J = \emptyset \},$$0). Formally,

$$\Pi^J = \{ H(r)^+ \leftarrow B(r)^+ \mid r \in \Pi, B(r)^- \subseteq J, B(r)^- \cap J = \emptyset \},$$1

with $$\Pi^J = \{ H(r)^+ \leftarrow B(r)^+ \mid r \in \Pi, B(r)^- \subseteq J, B(r)^- \cap J = \emptyset \},$$2. This operator requires determining all minimal changes, introducing computational complexity (Slota et al., 2013).

5. Computational Complexity and Limitations

Deciding, for the Winslett-style SE update operator $$\Pi^J = \{ H(r)^+ \leftarrow B(r)^+ \mid r \in \Pi, B(r)^- \subseteq J, B(r)^- \cap J = \emptyset \},$$3, the entailment $$\Pi^J = \{ H(r)^+ \leftarrow B(r)^+ \mid r \in \Pi, B(r)^- \subseteq J, B(r)^- \cap J = \emptyset \},$$4 (i.e., whether every SE-model of the update is an SE-model of $$\Pi^J = \{ H(r)^+ \leftarrow B(r)^+ \mid r \in \Pi, B(r)^- \subseteq J, B(r)^- \cap J = \emptyset \},$$5) is $$\Pi^J = \{ H(r)^+ \leftarrow B(r)^+ \mid r \in \Pi, B(r)^- \subseteq J, B(r)^- \cap J = \emptyset \},$$6-complete in the general case, even for positive facts and non-disjunctive $$\Pi^J = \{ H(r)^+ \leftarrow B(r)^+ \mid r \in \Pi, B(r)^- \subseteq J, B(r)^- \cap J = \emptyset \},$$7 and single fact $$\Pi^J = \{ H(r)^+ \leftarrow B(r)^+ \mid r \in \Pi, B(r)^- \subseteq J, B(r)^- \cap J = \emptyset \},$$8. For definite (Horn) programs, the problem is co-NP-complete (Slota et al., 2013).

A fundamental issue arises for all SE-model based update operators satisfying strong syntax-independence (P4). They must violate at least one of two basic properties:

1. Dynamic support: Every atom $$\Pi^J = \{ H(r)^+ \leftarrow B(r)^+ \mid r \in \Pi, B(r)^- \subseteq J, B(r)^- \cap J = \emptyset \},$$9 in an answer set after update should be supported by some rule in the union of original and update programs.
2. Fact update: Updates between consistent sets of facts should reflect database-style literal inertia.

An explicit example shows this violation: two programs with identical SE-models, but different syntactic support for an atom in the answer set. Any fully semantic, AGM/KM–style update operator on SE-models necessarily loses either support or the fact update property (Slota et al., 2013).

6. SE-Model Induced Notions of Program Equivalence

In addition to strong equivalence ($H(r) = (H(r)^+,H(r)^-)$0) and strong update equivalence ($H(r) = (H(r)^+,H(r)^-)$1), SE-model semantics give rise to further notions:

• Strong Rule equivalence ($H(r) = (H(r)^+,H(r)^-)$2): Two programs are SR-equivalent if their rule SE-model sets coincide (modulo tautology).
• Strong Minimal Rule equivalence ($H(r) = (H(r)^+,H(r)^-)$3): Minimal elements (under set inclusion) of their rule SE-model sets coincide.

These form a strict refinement hierarchy: $H(r) = (H(r)^+,H(r)^-)$4. Concrete examples distinguish between these, highlighting that the traditional strong equivalence is strictly weaker than the distinctions available when the full SE-model structure of rules is considered (Slota et al., 2011).

Equivalence Type	Condition	Comparative Strength
Strong Update ($H(r) = (H(r)^+,H(r)^-)$5)	SE-tautological difference in updates	Strongest
Strong Rule ($H(r) = (H(r)^+,H(r)^-)$6)	Same sets of rule SE-models
Strong Minimal Rule ($H(r) = (H(r)^+,H(r)^-)$7)	Same minimal rule SE-model sets
Strong Equivalence ($H(r) = (H(r)^+,H(r)^-)$8)	Identical SE-model sets	Weakest

7. Significance and Limitations of the SE-Model Semantic Approach

SE-models enable a uniform, monotonic, and semantic framework extending stable model semantics to characterize strong equivalence, program updates, and rule expressivity in non-monotonic logic programming. They provide a foundation for lifting classical belief-change and update methodologies (notably AGM and KM frameworks) into the non-monotonic setting, with representation theorems precisely characterizing the space of update operators that satisfy semantic postulates.

However, SE-model–based approaches, in their pure semantic and syntax-independent versions, necessarily lose certain pragmatic properties fundamental in answer-set programming, such as dynamic support and the realistic handling of fact updates. This suggests that to restore such properties, either more expressive semantic frameworks (beyond Here-and-There) or approaches re-integrating syntactic information are required (Slota et al., 2013). The transition from program syntax to pure semantic content, while theoretically clean, entails unavoidable tradeoffs in the behavior of updates and equivalence notions.

• "The Rise and Fall of Semantic Rule Updates Based on SE-Models" (Slota et al., 2013)
• "Back and Forth Between Rules and SE-Models (Extended Version)" (Slota et al., 2011)