TOPICS
Search

Upper Copula Capacity


The upper copula capacity of a Borel set A subset= [0,1]^n is the supremum of the probabilities assignable to A by choosing the dependence between random variables whose marginal distributions are uniform on [0,1]. Writing C_n for the corresponding set of copula probability measures, it is

 C^__n(A)=sup_(C in C_n)C(A).
(1)

The distinction between a copula measure and its joint distribution function is understood (Nelsen 2006, Côté and Wang 2026). The supremum is attained for closed A, but need not be attained for general Borel sets. This is a monotone, normalized, subadditive set function; it need not be a measure or a submodular set function. Componentwise transformations preserving the uniform distribution preserve its value under preimages (Côté and Wang 2026).

For a nonempty convex set A subset= [0,1]^2, let a_i and b_i be the infimum and supremum of its ith coordinate projection. Then

 C^__2(A)=min{b_1-a_1,b_2-a_2}.
(2)

This two-dimensional formula fails in higher dimensions. The regular octahedron {u in [0,1]^3:sum_(i)|u_i-1/2|<=1/2}, for example, has full coordinate projections but upper copula capacity 2/3 (Côté and Wang 2026).

For positive weights lambda_1,...,lambda_n and s>=0, the exact lower-tail formula is

 C^__n({u in [0,1]^n:sum_(i=1)^nlambda_iu_i<=s})=min(s/L,1),
(3)

where L=max{max_(i)lambda_i,(1/2)sum_(i)lambda_i} (Côté and Wang 2026). For a decreasing Borel set A, meaning that u in A and 0<=v<=u coordinatewise imply v in A, one also has C^__n(tA)=tC^__n(A) for 0<=t<=1.

For a bounded nonnegative upper semicontinuous function psi, its upper Choquet integral is J(psi)=int_0^inftyC^__n(psi>t)dt. It is at least sup_(C in C_n)intpsidC. Equality holds exactly when one copula simultaneously attains C(psi>t)=C^__n(psi>t) for every real t (Côté and Wang 2026). Thus integration against the upper capacity can exceed the largest expectation under any single copula.

A finite combinatorial interpretation uses n vertex classes of size m and a multipartite hypergraph. Partition each coordinate interval into m equal parts and let A be the union of boxes corresponding to the hyperedges. Then mC^__n(A) is the maximum fractional matching size, where weights on hyperedges are nonnegative and sum to at most 1 at each vertex (Côté and Wang 2026).


See also

Copula, Marginal Distribution, Optimal Transport

Explore with Wolfram|Alpha

References

Côté, B. and Wang, R. "The Upper Copula Capacity." 18 Nov 2026. https://arxiv.org/abs/2611.25237v1.Nelsen, R. B. An Introduction to Copulas, 2nd ed. New York: Springer, 2006. https://doi.org/10.1007/0-387-28678-0.

Cite this as:

Weisstein, Eric W. "Upper Copula Capacity." From MathWorld--A Wolfram Resource. https://mathworld.wolfram.com/UpperCopulaCapacity.html

Subject classifications