Generalized Boole–Fréchet Bounds

• Generalized Boole–Fréchet bounds are sharp probabilistic limits that extend classical union and intersection inequalities to complex, multidimensional scenarios.
• They employ combinatorial identities, moment-based aggregations, and optimization techniques like dynamic programming to tighten bounds under partial information.
• Applications span reliability theory, contingency table analysis, risk aggregation, and quantum probability, demonstrating their practical impact across diverse fields.

A generalized Boole–Fréchet bound is any sharp probabilistic bound on the measure of joint or compound events under given marginal measures and possibly additional partial information, generalizing the classical Boole union and Fréchet intersection inequalities. Generalized bounds have been developed for Boolean events, multivariate random vectors, contingency tables, compound symmetric events, copulas, and quantum observables, yielding a rich hierarchy of theoretical results and efficient computational techniques with deep implications for statistics, risk, reliability, imprecise probability, and mathematical optimization.

1. Classical Boole–Fréchet Inequalities and Their Generalization

The classical Boole and Fréchet inequalities provide extremal lower and upper bounds for $\mathbb P(\bigcap_{i=1}^n A_i)$ and $\mathbb P(\bigcup_{i=1}^n A_i)$ given only marginal probabilities $p_i = \mathbb P(A_i)$. For example, with $n$ events,

$$\max\{0,\sum_{i=1}^n p_i - (n-1)\} \leq \mathbb P(\bigcap_{i=1}^n A_i) \leq \min_i p_i, \quad \mathbb P(\bigcup_{i=1}^n A_i) \leq \sum_{i=1}^n p_i.$$

The Bonferroni inequalities—derived by truncating inclusion–exclusion at an appropriate order—systematically refine these bounds via additional joint moments, exhibiting monotonic improvement with increasing combinatorial order (Salako, 9 Dec 2025, Ding et al., 2015).

Generalized Boole–Fréchet bounds extend these results in multiple directions:

• Compound and order-statistics events: E.g., for the event that at least $k$ out of $n$ events occur, sharp bounds can be constructed via dynamic programming, exactly attaining the extremal value (Salako, 9 Dec 2025).
• Joint distributions with higher-order marginals: Optimizing $P(\bigcup_{i=1}^n A_i)$ or $P(\bigcap_{i=1}^n A_i)$ with additional knowledge of pairwise or higher joint measures leads to tightening of classical bounds, realized formally as a high-dimensional linear programming problem (Hailperin model, cf. (Boros et al., 2021)).
• Multivariate bounds: For vectors or contingency tables, framing bounds in terms of multivariate binomial moments or lattice marginalization yields explicit, improved inequalities (Uhler et al., 2017, Ding et al., 2015).
• Bounds for copulas under partial information: In the setting of multivariate dependence structures, knowing the copula at certain points, or the value of a concordance-monotone functional (e.g., Kendall’s $\tau$), allows strict improvements over the Fréchet–Hoeffding bounds (Tankov, 2010).
• Quantum generalization: For measurements on noncommuting observables, Boole–Fréchet bounds acquire quantum correction terms dependent on noncommutativity; these correct for state-dependent violations by entangled states (Vourdas, 2019).

2. Combinatorial and Moment-Based Generalizations

Key advances in generalized Boole–Fréchet bounds leverage combinatorial identities and moment calculations:

• For integer-valued random pairs $(S,T)$, joint tail probabilities $P(S\ge u, T\ge v)$ are expanded exactly in terms of bivariate binomial moments $S_{i,j} = E[\binom{S}{i}\binom{T}{j}]$ via an alternating double sum (Ding et al., 2015):

$$P_{u,v} = \sum_{i=u}^m \sum_{j=v}^n (-1)^{(i-u)+(j-v)} \binom{i-1}{u-1} \binom{j-1}{v-1} S_{i,j}.$$

• Truncating these sums yields Gumbel–Fréchet–type upper/lower Bonferroni bounds, systematically improving as higher moments are incorporated. These methods highlight the multiplicative and inclusion–exclusion structures underlying generalized bounds.
• In the hierarchy of bounds on $P(\cup_i A_i)$, aggregations of atomic probabilities into moment-type summaries (binomial moments, size-and-node moments, quadratic moments) yield polyhedral relaxations of the intractable original LP model—each step corresponding to a boundary in the hierarchy of known Bonferroni, Prékopa–Gao, and cut polytope–based bounds (Boros et al., 2021).

3. Multivariate and Lattice-Based Extensions

For contingency tables in $\ell$ dimensions, the marginalization function over lattice $\mathcal L = 2^L$ ($L = \{1,\dots,\ell\}$) is central. The function $n_A(x_A)$—the sum over entries with indices reducing to $x_A$ on $A \subseteq L$—is entrywise monotonic and supermodular:

• Monotonicity: $n_A(x_A) \geq n_B(x_B)$ for $A \subseteq B$.
• Supermodularity: $$n_{A\cup B}(x_{A\cup B}) + n_{A\cap B}(x_{A\cap B}) \geq n_A(x_A) + n_B(x_B)$$.

These structural properties yield, for any two marginals $A,B$,

$$\min \{ n_A(x_A), n_B(x_B) \} \geq n_{A\cup B}(x_{A\cup B}) \geq \max\{ n_A(x_A) + n_B(x_B) - n_{A\cap B}(x_{A\cap B}), 0 \},$$

a direct multidimensional analogue of Boole–Fréchet bounds (Uhler et al., 2017). Extensions by Dobra–Fienberg inequalities and Ky Fan supermodular inequalities further generalize these to arbitrary collections of marginal constraints on the lattice.

Ky Fan’s inequality, applied to supermodular functions on distributive lattices, furnishes a mechanism for deriving tight generalizations of Boole–Fréchet bounds for joint probabilities determined by arbitrary collections of marginals.

4. Constructive and Optimization-Based Methodologies

Generalized Boole–Fréchet bounds are often computationally realized as sharp solutions to high-dimensional linear programs or dynamic programs:

• Hailperin model: Considers the atomic decomposition of probability space into $2^n-1$ outcomes, imposing equality constraints corresponding to known marginals and optimizing the desired event’s probability. The extremal value is the optimal value of an exponentially sized LP (Boros et al., 2021, Salako, 9 Dec 2025).
• Aggregation hierarchies: Moment-based and combinatorial aggregations (binomial, size-and-node, quadratic) yield reduced polyhedral cones and corresponding LP relaxations. Adding facet-defining inequalities from, e.g., the cut polytope, tightens these relaxations, unifying classical probabilistic bounds with combinatorial optimization (Boros et al., 2021).
• Dynamic programming: Especially for symmetric or “$k$-out-of-$n$” event classes, dynamic programming exhibits constructive Bellman recursions yielding tight bounds, with analytic expressions in some special cases (Salako, 9 Dec 2025).

Simple cases allow explicit analytic solutions; for “at least $k$ out of $n$” event,

$$\sup_{\mathbb P(A_i) = p_i} \mathbb P\bigl(|\{i : A_i\}| \geq k \bigr) = \min\left\{ \frac{p_1 + \cdots + p_{n - r^*}}{k - r^*},\, 1 \right\},$$

where $r^*$ is determined by the ordering of marginals (Salako, 9 Dec 2025).

5. Copula-Theoretic and Dependence Information Generalizations

Generalized Boole–Fréchet bounds for joint distributions with known marginals but only partial information about dependencies have been characterized via improvements to the Fréchet–Hoeffding–type bounds on copulas:

• If the copula $C$ of a bivariate random vector is known on a subset $S \subset [0,1]^2$ (or for a concordance-monotone functional $\rho(C)$), improved bounds are given by quasi-copula envelopes that refine $$W(u,v) = \max\{0, u + v - 1\} \leq C(u,v) \leq \min\{u,v\} = M(u,v)$$ (Tankov, 2010).
• For higher dimensions, the bounds are given by

$$W_d(u) = \max\left\{0, \sum_{i=1}^d u_i - (d-1)\right\}, \quad M_d(u) = \min_{1 \leq i \leq d} u_i,$$

with analogous envelope constructions exploiting known values or functional constraints.

This copula-theoretic generalization is critical in quantitative finance (model-free option pricing), risk management, and robust statistics.

6. Quantum Probabilistic Extensions

In quantum probability, Boole–Fréchet inequalities acquire non-classical correction terms when events correspond to noncommuting observables. Defining projection operators $\Pi(h)$ for subspaces $h$, the quantum correction operator $\mathfrak D(h_1, h_2)$ captures non-commutative effects: $p(h_1 \vee h_2) \leq p_1 + p_2 + \Delta,$ where $$\Delta = \langle s | \mathfrak D(h_1, h_2) | s \rangle$$. The classical inequalities are recovered when projection operators commute or the state is factorizable; quantum corrections are necessary for entangled or nonclassical states, as in some violations of the CHSH inequality (Vourdas, 2019).

7. Applications and Impact Across Disciplines

Generalized Boole–Fréchet bounds are indispensable in:

• Reliability theory: Computing joint failure probabilities with only marginal failure probabilities known or partially estimated.
• Risk aggregation in finance and insurance: Bounding loss probabilities under imprecise dependence constraints between risk factors.
• Contingency table analysis: Bounding cell entries given constrained marginals in survey sampling, disclosure limitation, and ecological inference (Uhler et al., 2017).
• Statistics and AI: Inference under imprecise or partial probabilistic information, robust uncertainty quantification.
• Quantum information: Verifying classical and quantum probabilistic inequalities in measurement scenarios.

The impact of these bounds extends to any domain where only partial joint probability information is available but sharp, interpretable probabilistic bounds are critical.

---

References:

• (Salako, 9 Dec 2025) Constructive Proofs of Generalized Boole--Frechet Bounds: A Dynamic Programming Approach
• (Boros et al., 2021) Boole's probability bounding problem, linear programming aggregations, and nonnegative quadratic pseudo-Boolean functions
• (Ding et al., 2015) Bivariate Binomial Moments and Bonferroni-type Inequalities
• (Uhler et al., 2017) Generalized Fréchet Bounds for Cell Entries in Multidimensional Contingency Tables
• (Tankov, 2010) Improved Frechet bounds and model-free pricing of multi-asset options
• (Vourdas, 2019) Probabilistic inequalities and measurements in bipartite systems