Logic of Preference and Functional Dependence

• LPFD is a formal system integrating functional dependence, preference modalities, and coalitional reasoning to model strategic and collective decision-making.
• It employs a rich syntax of atomic predicates, functional dependence atoms, and group-preference operators to express game-theoretic equilibria such as Nash and Pareto optimality.
• The logic supports decidable reasoning with a Hilbert-style proof system and PSPACE-complete model-checking, enabling applications in databases, multi-agent systems, and cooperative game theory.

The Logic of Preference and Functional Dependence (LPFD) is a formal system integrating functional dependence, preference modalities, and coalitional reasoning. It provides a unified framework to reason about information flow, agent preferences, and strategic interaction, facilitating the expression and analysis of concepts such as Nash equilibrium, Pareto optimality, and the economic core in multi-agent systems. LPFD generalizes classical logics of functional dependence by augmenting them with a rich structure of preference modalities, and extends to hybrid logics to model coalitional power and collective agency (Shi et al., 2021, Chen et al., 2022).

1. Syntax and Language Construction

LPFD is built over a countable set of “players” $V$, a relational vocabulary $\mathrm{Pred}$, and (for hybrid extensions) a set of nominals. Formulas are constructed from:

• Atomic predicates: $P\, x_1 \ldots x_n$, for $P$ in $\mathrm{Pred}$.
• Functional dependence atoms: $D_X\, y$ asserts that $y$'s action is functionally determined by the actions of $X$ ($X \subseteq V$, $y \in V$).
• Boolean connectives: Negation ($\neg$), conjunction ($\wedge$).
• Group-preference (ceteris paribus) operators: $[X,Y,Z]\, \varphi$, where $X,Y,Z \subseteq_{\mathrm{fin}} V$, governed by agreement (on $X$), weak preference ($Y$), and strict preference ($Z$).
• Hybrid operators: (in HLPFD) nominals $i$ and satisfaction operators $@_i\, \varphi$.

Abbreviations such as $$\mathbb{D}_X\, \varphi := [X,\emptyset,\emptyset]\varphi$$ generalize familiar universal modalities from modal logic. The ceteris paribus modalities $$\langle X,Y,Z\rangle\,\varphi := \neg [X,Y,Z]\,\neg \varphi$$ allow existential preference-based reasoning.

This expressive language subsumes both classical (monoidal) functional dependency expressions and the full spectrum of preference and group-modality operators required for strategic game-theoretical and agency reasoning (Shi et al., 2021, Chen et al., 2022, Vychodil, 2014).

2. Semantic Models

An LPFD semantics is based on a preference-dependence model (PD-model):

• Agents and choices: $O$ is a nonempty set of "choices" (actions). $A \subseteq O^V$ is the set of admissible strategy profiles, $s: V \to O$.
• Predicates: $I(P) \subseteq O^{\operatorname{ar}(P)}$ gives the interpretation of each $n$-ary predicate $P$.
• Preference relations: For each player $x \in V$, a reflexive, transitive preorder $\preceq_x$ and its strict part $\prec_x$ on $A$ specify individual preferences over global profiles.

Truth at $a \in A$ is defined inductively:

• $D_X y$ holds at $a$ iff all $a' \in A$ with $a' =_X a$ satisfy $a'(y) = a(y)$.
• $[X,Y,Z]\varphi$ holds at $a$ iff for every $a' \in A$ with $a' =_X a$, $a \preceq_Y a'$, and $a \prec_Z a'$, we have $M, a' \models \varphi$.

Kripke-style relational semantics are equivalent, where $[X,Y,Z]$ is interpreted using tuples of equivalence and preference relations indexed by players.

This flexible semantic base enables both propositional dependence and classical relational "database-style" readings (as in monoidal functional dependencies (Vychodil, 2014)), as well as the modeling of game-theoretic preference structures (Chen et al., 2022, Shi et al., 2021).

3. Proof Systems and Meta-theorems

LPFD is axiomatized by a Hilbert-style calculus $\mathcal{C}_{\mathrm{LPFD}}$ with:

• All propositional tautologies and Modus Ponens.
• Normal modal/necessitation rules for $[X,Y,Z]$.
• Axioms for $[X,Y,Z]$: including reflexivity, distribution, monotonicity, combination, and ceteris paribus interaction principles, allowing one to reason systematically about various levels of group agreement and preference.
• Axioms for $D_X$ (functional dependence): including reflexivity, atomic coverage, transitivity, and dependence-propagation under group-preference.
• Hybrid logic axioms (for HLPFD): support for nominals, named points, and satisfaction operators, critical for formal coalitional power (Chen et al., 2022).

Principal Meta-theorems

• Soundness: All derivable formulas are semantically valid in all PD-models.
• Strong Completeness: For any set $\Sigma$ and formula $\varphi$, if $\Sigma \models \varphi$ then $\Sigma \vdash_{\mathrm{LPFD}} \varphi$.
• Decidability: Despite lacking the finite-model property, LPFD has a decidable satisfiability problem; a filtration argument suffices to obtain a decision procedure (Chen et al., 2022). For fixed finite models, model-checking is in PSPACE (Shi et al., 2021).
• Complexity: The exact complexity of general satisfiability remains open, but at least PSPACE-hardness is expected.

4. Expressive Power: Functional Dependence, Nash, Pareto, and the Core

LPFD internalizes standard game-theoretic solution concepts:

• Nash equilibrium for $X \subseteq V$:

$$\operatorname{Na}\, X := \bigwedge_{x \in X} [V \setminus\{x\}, \emptyset, \{x\}]\, \bot$$

expressing that no $x \in X$ can strictly improve by unilaterally changing strategy.

• Strong Pareto optimality for $X$:

$$\operatorname{sPa}\, X := \bigwedge_{x \in X} [V \setminus X, X \setminus \{x\}, \{x\}]\, \bot$$

asserting that it is impossible to strictly improve the lot of some agent in $X$ without worsening another when holding $V \setminus X$ fixed.

• Weak Pareto optimality:

$$\operatorname{wPa}\, X := [V \setminus X, \emptyset, X]\, \bot$$

• The core (in HLPFD, for coalitional games): Via coalitional-supporting abbreviations, the formula

$$\mathrm{Core}\, i := i \wedge p_N \wedge \bigwedge_{\emptyset \ne X \subset N} (p_X \rightarrow [X,\emptyset,\emptyset] \bigvee_{x \in X} [\emptyset,\{x\},\emptyset]\, i)$$

names the core as those profiles unblocked by strict subcoalitions (Chen et al., 2022).

These encodings permit direct modal-logical analysis and agent-based reasoning within a single formal system (Shi et al., 2021, Chen et al., 2022).

5. Connections and Relations to Other Logics

LPFD generalizes several logical frameworks:

• Coalition Logic: Pauly's coalition operator $[C]\varphi$ corresponds to LPFD's $$\mathbb{D}_C\, \varphi = [C,\emptyset,\emptyset]\varphi$$. LPFD reveals the role of functional dependence in blocking “superadditivity” in general settings without full support $A = O^V$.
• Ceteris Paribus Preference Logic: Van Benthem–Hansson's $[\Gamma]\varphi$ is mapped to $[\Gamma,\emptyset,\emptyset]\varphi$, but LPFD introduces parameterization by agreement, weak/strict preference, and functional dependence.
• Dependence Logic (Väänänen) and Modal LFD: LPFD extends modal logic of functional dependence (LFD) with preference preorders, resulting in a “hub-and-spoke” architecture for dependence, preference, and coalitional modalities (Chen et al., 2022).
• Monoidal Functional Dependencies (MFD): MFDs in the sense of Vychodil (Vychodil, 2014) correspond to functional dependencies with similarity-based semantics in commutative integral pomonoids and residuated lattices.

Table: Correspondence of Key Modal Constructs

Concept	CP/Coalition Logic	LPFD Expression
Coalition ability	$[C]\varphi$	$[C, \emptyset, \emptyset]\varphi$
Ceteris paribus preference	$[\Gamma]\varphi$	$[\Gamma, \emptyset, \emptyset]\varphi$
Functional dependence	$D_X y$	$D_X y$
Nash equilibrium (player $x$)	—	$[V \setminus\{x\}, \emptyset, \{x\}]\, \bot$

LPFD not only subsumes preceding logics but provides uniform methodology for integrating dependence, preference, and agency under coalitional structures.

6. Illustrative Examples and Applications

In classical database scenarios, MFDs (the proto-LPFD logic) enable nuanced dependencies, such as expressing that "similar LOCATION and AREA imply similar PRICE" by $$(\mathrm{LOCATION}\,\mathrm{AREA}) \Rightarrow \mathrm{PRICE}$$ under evaluations in $[0,1]$-valued similarity structures (Vychodil, 2014). In game-theoretic settings, the “Rock–Jazz” coordination game exemplifies how LPFD can model Nash equilibria, Pareto optimal outcomes, and the effect of functional dependencies on strategic stability (Shi et al., 2021).

In cooperative game theory, LPFD and its hybrid extension HLPFD allow one to characterize the core of a coalition game, capturing collective agency as a “stability” condition—profiles from which no strict sub-coalition can deviate to improve its members' welfare (Chen et al., 2022).

The framework is thus applicable wherever complex dependencies and agent preferences interact: database dependency theory, multi-agent systems, collective agency in philosophy, and the core concepts of modern game theory.

7. Decidability and Computational Aspects

LPFD's satisfiability problem is decidable, with the decision procedure based on filtration constructions over the finitely many subformulas of bounded modal depth. Despite lacking the finite-model property, the logic supports effective reasoning by finite construction. Model checking for a given finite PD-model is in PSPACE, with at least PSPACE-hardness for satisfiability. In the monoidal setting, fragments restricted to "non-contracting" dependencies admit polynomial-time algorithms generalizing those used for standard functional dependencies (Chen et al., 2022, Shi et al., 2021, Vychodil, 2014).

A plausible implication is that, despite the high expressive power, LPFD offers feasible avenues for automated reasoning in multi-agent, database, and game-theoretic applications, especially for logically structured preference and dependency constraints.