Quantitative Predicate Logic (QPL)

• Quantitative Predicate Logic (QPL) is a logic that integrates real-valued semantics using power means, probabilistic interpretations, and aggregative quantifiers.
• It introduces multiple classes of connectives—nonlinear, linear-additive, and linear-multiplicative—to capture diverse inference modes from softmax to argmax.
• QPL bridges classical deduction and quantitative reasoning, providing foundations for decision procedures, entropy measures, and diversity indices.

Quantitative Predicate Logic (QPL) generalizes classical predicate logic by introducing real-valued, probabilistic, counting, and aggregative semantics for predicates, quantifiers, and inference. QPL aims to reason not only about the truth or falsity of statements, but also about the degree, count, or measure to which such statements hold. Theoretical foundations span lattice- and quantale-enriched logics, higher-order continuous semantics, logical frameworks for counting, and practical decision mechanisms reducible to systems of linear inequalities or probabilistic interpretations.

1. Core Syntax and Semantics of Quantitative Predicate Logic

QPL extends first-order predicate logic by allowing formulas whose interpretations are not limited to $\{0,1\}$ but instead take values in domains such as $[0,\infty]$, $[-\infty,\infty]$, $[0,1]$, or $\mathbb{N}$. In the formalism developed in "On Quantifiers for Quantitative Reasoning" (Capucci, 2024), two principal languages are simultaneously formed over each measured space $I$:

• Multiplicative-real QPL: Atomic predicates are measurable functions $I \xrightarrow{a} [0, \infty]$. Formulas are built from constants, these atomic predicates, six binary connectives (nonlinear, linear-additive, and linear-multiplicative), involutive duality (negation as $x \mapsto 1/x$), scalar modalities, and power-means-based quantifiers:

$$\varphi, \psi ::= c \mid a(i) \mid \varphi\,\mathsf{op}\,\psi \mid \varphi \multimap \psi \mid \varphi^* \mid k\cdot\varphi \mid \exists^p_{j \in J} \psi(i,j) \mid \forall^p_{j \in J} \psi(i,j)$$

• Additive-real QPL: Syntax as above, now with constants in $[-\infty, \infty]$, and duality $x \mapsto -x$. Translations between the two are given by $-\log \dashv 1/\exp$.

Quantifiers generalize existential and universal quantification via $p$-means (power means), interpolating between min, max, arithmetic mean, and geometric mean depending on $p$. For discrete $x_1, ..., x_n$:

$$\mathrm{M}_p(x_1,...,x_n) = \left(\frac{1}{n}\sum_{i=1}^n x_i^p\right)^{1/p}$$

with $p\to+\infty$ yielding $\max$, $p\to-\infty$ giving $\min$, and $p=1,0$ corresponding to arithmetic and geometric mean, respectively.

In measurable contexts, these quantifiers take integral forms:

$$\left(\exists^p_{j \in J} \psi\right)(i) = \left(\int_{j \in J} \psi(i,j)^p\, dj\right)^{1/p} \qquad \left(\forall^p_{j \in J} \psi\right)(i) = \left(\int_{j \in J} \psi(i,j)^{-p}\, dj\right)^{-1/p}$$

This framework supports soft and classical quantifiers in a uniform parametric continuum and extends to additive semantics via the logarithmic transformation (Capucci, 2024).

2. Generations of Connectives: Nonlinear, Additive, and Multiplicative

QPL admits three interrelated "generations" or classes of real-valued binary connectives, each with distinct algebraic structure and logical behavior (Capucci, 2024):

• Nonlinear (idempotent):
  • Join ("or"): $x \lor y = \max\{x, y\}$, unit $0$
  • Meet ("and"): $x \land y = \min\{x, y\}$, unit $\infty$
• Linear-additive:
  • Tensor: $x \oplus y := x + y$, unit $0$
  • Cotensor (De Morgan dual): $x \ominus y := (1/x+1/y)^{-1}$ ("harmonic sum"), unit $\infty$
• Linear-multiplicative:
  • Tensor: $x \otimes y := x \cdot y$, unit $1$ ($0\cdot\infty := 0$)
  • Cotensor: $x \parr y := x\cdot y$ (reverse units: $0\cdot\infty = \infty$, $\bot=1$)

These connectives distribute over each other in accordance with lax-linear distributivity. The involutive duality $x \mapsto 1/x$ exchanges additive and multiplicative connectives. Using the $-\log \dashv 1/\exp$ duality, these structures are transferred to the additive world: sums become log-sum-exp, products become additive, and so on.

This multiplicity of connectives supports the expressive integration of min/max, means, softplus, and product-based reasoning in a single logic.

3. Quantifiers via Power Means and Quantitative Aggregation

Quantitative generalization of existential and universal quantifiers is achieved using power means (and their harmonic variants), leading to "soft" quantifiers that interpolate between standard $\max$, $\min$, and averaging behaviors (Capucci, 2024):

• Discrete $p$-quantifiers:

$\bigoplus^p_{i=1}^n x_i := \left(\sum_{i=1}^n x_i^p\right)^{1/p}$

$\bigominus^p_{i=1}^n x_i := \left(\sum_{i=1}^n x_i^{-p}\right)^{-1/p}$

• Continuous (measurable) $p$-quantifiers:

$$(\exists^p_{j \in J}\psi)(i) = \Bigl(\int \psi(i,j)^p\,dj\Bigr)^{1/p}$$

$$(\forall^p_{j \in J}\psi)(i) = \Bigl(\int \psi(i,j)^{-p}\, dj\Bigr)^{-1/p}$$

As $p \to +\infty$, these quantifiers retrieve $\max$ or existential quantification; as $p \to -\infty$, they retrieve $\min$ or universal quantification. For $p=1$, arithmetic mean arises; for $p=0$, geometric mean.

The additive-world forms, via $-\log$ transformation, correspond to

$$(\exists^p \psi)(i) = -\frac{1}{p} \log \int_j e^{-p\psi(i,j)}\, dj$$

This construction accommodates a continuum of logical and statistical aggregation mechanisms, enabling expressive power for graded inference.

4. Illustrative Applications: Softmax, Argmax, Entropy, and Hill Numbers

Quantitative semantics of QPL connect directly to major constructions in analysis, statistics, and information theory (Capucci, 2024):

• Argmax vs. Softmax:
  • Classic argmax is realized via a Boolean universal quantification.
  • In QPL, soft quantification yields softmax as a normalized, $p=1$ instance:

  $$\mathrm{softmax}(f)(x^*) = \frac{f(x^*)}{\sum_{x \in X} f(x)}$$ - As $p \to \infty$, the "soft" quantification recovers the strict "argmax".

• Rényi Entropy and Hill Numbers:
  • Rényi entropy of order $p$ for a density $\varphi$:

  $$H_p(\varphi) = \frac{1}{1-p}\log \int \varphi(i)^p\, di$$ - This is an additive QPL soft-universal of $-\log \varphi$. - The multiplicative QPL dual computes Hill numbers (diversity indices), showing the duality:

  $$\frac{p}{1-p} \bigl(\forall^p_i\, \varphi(i)^*\bigr)$$ - The Napierian duality $-\log \dashv 1/\exp$ thus explains the interrelation between additive entropy and multiplicative diversity measures, highlighting a structural symmetry between probabilistic and energetic viewpoints in logic.

5. Decision and Model-Theoretic Properties

QPL, in forms associated with counting quantifiers and extended constraints (CQUEL fragment), admits exact characterizations linking logical satisfiability to solvability of systems of (integer) linear inequalities (Finger, 2019):

• Each satisfiable QPL formula (counting fragment) has a finite model.
• Satisfiability is NP-complete, and the decision process can be framed as a column-generation plus branch-and-bound procedure over integer linear programs.
• Every model can be witnessed with a small (logarithmic in arity) number of nonzero types.
• QPL coalesces with propositional probabilistic satisfiability (PSAT) and infinitely-valued probabilistic logic (LIPSAT) in the spirit of reasoning systems blending deductive and quantitative inference.

For power-mean-based and metric-based frameworks, the logical consequence notions, completeness, and soundness are tightly connected to the algebraic and analytic properties of the underlying real-valued lattices or metric spaces.

6. Categorical Semantics and Challenges

Attempts to host QPL within categorical frameworks such as Lawvere-style hyperdoctrines encounter structural obstacles (Capucci, 2024):

• The natural requirements of identity and cut rules, stemming from Boolean/idempotent behavior, are generally violated by soft quantifiers in QPL.
• Naïve harmonic-mean-based entailment fails reflexivity and transitivity except on probability spaces.
• More sophisticated enriched categorical approaches collapse because pushforward operations do not support requisite monoidal structures.
• These failures point to a structural mismatch between idempotent logical meta-levels and non-idempotent quantitative inference. A plausible implication is the need for the development of genuinely "quantitative" analogues of logical hyperdoctrines, tailored to support graded, non-idempotent inference.

7. Broader Landscape and Related Frameworks

QPL is a unifying formalism for systems that integrate aggregation, counting, probability, and degrees of truth within logical inference. It provides a foundation for:

• Real-valued logics enriched with means, entropies, and statistical functionals (Capucci, 2024)
• Counting-quantifier logics with reduction to integer linear programs (Finger, 2019)
• Lambda-calculus-based, algorithmically-precise quantifiers including counting and sum ("Σ") (Shkotin, 2019)
• Quantitative generalizations of conditional logic and security protocol reasoning, where proof systems handle explicit error and confidence parameters (0804.2155)
• Metric, affine, and higher-order QPLs for reasoning about probabilistic programs and distances (Bacci et al., 30 Jan 2025)

This ecosystem supports a range of applications from formal verification to knowledge representation, and points toward further inquiry in categorical semantics, expressivity over infinite domains, and algorithmic tractability under varying semantic conventions.