Associated Kernels

• Associated kernels are specialized functions constructed from prior structures—like boundaries, semigroups, or semimetrics—to retain geometric and algebraic properties.
• They are used across potential theory, RKHS, and nonparametric statistics to address issues like support mismatch and boundary leakage.
• Mechanisms such as balayage, symmetry invariance, and negative-type adjustments ensure that the derived kernels are better suited for targeted estimation and analysis.

Associated kernels are kernels constructed from prior structure—another kernel, a boundary, a semimetric, a support constraint, a semigroup action, or a multiplier—so that the derived kernel retains and reorganizes the geometry, potential theory, or algebra of the source object. The literature represented here uses the term in several non-equivalent ways. This suggests a family-resemblance notion rather than a single formal definition: in potential theory it denotes Green kernels obtained from Riesz kernels by balayage, in RKHS and dilation theory it denotes invariant or distance-induced kernels canonically attached to semigroup actions or negative-type semimetrics, and in nonparametric statistics it denotes target-adaptive smoothing kernels whose support follows the domain of the data (Fuglede et al., 2016, Ay et al., 2015, Sejdinovic et al., 2012, Somé et al., 2015).

1. Conceptual scope and recurrent construction patterns

Across the cited literatures, association is rarely arbitrary. It is typically produced by one of a small number of mechanisms: projection or sweeping onto a constraint set, covariance with respect to a symmetry, compensation by a boundary term, anchoring a semimetric to obtain a positive definite kernel, or parameterizing a family so that the kernel’s support and moments track the estimation point. In each case, the derived kernel is designed to preserve a structural principle already present in the ambient problem—maximum principles in potential theory, reproducing properties in RKHS theory, invariance in representation theory, or support fidelity in smoothing.

A recurrent technical theme is that the associated kernel is often better behaved for the target problem than the original object. The $\alpha$-Green kernel $g_D^\alpha$ isolates the part of the $\alpha$-Riesz interaction not absorbed by the complement $D^c$; distance-induced kernels convert negative-type semimetrics into PD kernels suitable for MMD and HSIC; invariant operator-valued kernels produce linearisations and reproducing kernel VE-spaces carrying $*$-representations; and support-adaptive associated kernels reduce boundary leakage and support mismatch in nonparametric estimation (Fuglede et al., 2016, Sejdinovic et al., 2012, Ay et al., 2015, Kokonendji et al., 2015).

The same phrase also has strict terminological limits. In some algebraic literatures, “kernel” means an abstract kernel or coupling of an extension rather than a function of two variables. That usage is explicitly distinguished from functional-analytic and statistical kernel notions, and it is therefore essential not to impose a single cross-disciplinary definition where the source literature does not provide one (Li, 2018).

2. Balayage-associated kernels in potential theory

In the most classical sense represented here, associated kernels arise from balayage of the $\alpha$-Riesz kernel

$$\kappa_\alpha(x,y)=|x-y|^{\alpha-n}, \qquad 0<\alpha\le 2,\ n\ge 3.$$

For a closed set $A\subset \mathbb{R}^n$ and a positive Radon measure $\mu$, sweeping produces a unique measure $\mu^A$ carried by $g_D^\alpha$0 such that

$g_D^\alpha$1

For finite-energy measures, $g_D^\alpha$2 is equivalently the orthogonal projection of $g_D^\alpha$3 onto the convex cone $g_D^\alpha$4; sweeping is symmetric in the sense that $g_D^\alpha$5; and for $g_D^\alpha$6 carried by $g_D^\alpha$7 one has the Cartan-type integral representation

$g_D^\alpha$8

This extends Cartan’s Newtonian balayage theory from $g_D^\alpha$9 to $\alpha$0 (Fuglede et al., 2016).

The associated $\alpha$1-Green kernel on a domain $\alpha$2 is then defined by

$\alpha$3

The compensating term is $\alpha$4-harmonic in $\alpha$5, agrees q.e. with $\alpha$6 on $\alpha$7, and encodes the effect of the boundary through sweeping. The resulting kernel is symmetric, lower semicontinuous on $\alpha$8, continuous off the diagonal, strictly positive, and $\alpha$9 on the diagonal. For an extendible measure $D^c$0 on $D^c$1,

$D^c$2

and, for compactly supported $D^c$3,

$D^c$4

The data describe this as measuring precisely the part of $D^c$5 “not seen from $D^c$6” (Fuglede et al., 2016).

The associated Green kernel inherits a full potential-theoretic apparatus. It satisfies the complete maximum principle, including domination and Frostman variants; it is strictly positive definite, hence its energy defines a norm; and it is consistent, so in combination with strict positive definiteness it is perfect. Consequences include strong completeness of the cone of positive finite-energy measures and the fact that the strong topology is finer than the induced vague topology. The corresponding capacity

$D^c$7

vanishes exactly when the $D^c$8-Riesz capacity does, and relatively closed sets $D^c$9 of finite $*$0-capacity admit a unique equilibrium measure $*$1 with

$*$2

and

$*$3

No regularity of $*$4 is required. This suggests that the association-via-balayage mechanism is not merely representational; it is the device that makes the full Green-kernel theory available for fractional $*$5 as well as for the Newtonian case (Fuglede et al., 2016).

3. Invariant, reproducing, and factorized associated kernels

A second major usage concerns kernels associated with symmetry and representation. For a $*$6-semigroup $*$7 acting on a set $*$8, an $*$9-valued kernel $\alpha$0 is invariant when

$\alpha$1

The central result is an equivalence: such a kernel is PSD and invariant if and only if it admits a $\alpha$2-invariant VE-space linearisation $\alpha$3 with

$\alpha$4

and if and only if there exists a minimal $\alpha$5-reproducing kernel VE-space carrying a $\alpha$6-representation $\alpha$7 satisfying

$\alpha$8

The framework is explicitly non-topological: it uses order and $\alpha$9-structures without requiring norms or continuity assumptions (Ay et al., 2015).

This perspective fits naturally with feature-space realizations of kernels in $$\kappa_\alpha(x,y)=|x-y|^{\alpha-n}, \qquad 0<\alpha\le 2,\ n\ge 3.$$0. A PD kernel $$\kappa_\alpha(x,y)=|x-y|^{\alpha-n}, \qquad 0<\alpha\le 2,\ n\ge 3.$$1 can be represented by a measurable field of features $$\kappa_\alpha(x,y)=|x-y|^{\alpha-n}, \qquad 0<\alpha\le 2,\ n\ge 3.$$2 such that

$$\kappa_\alpha(x,y)=|x-y|^{\alpha-n}, \qquad 0<\alpha\le 2,\ n\ge 3.$$3

The associated RKHS then embeds isometrically into the closed span of the features through

$$\kappa_\alpha(x,y)=|x-y|^{\alpha-n}, \qquad 0<\alpha\le 2,\ n\ge 3.$$4

with adjoint

$$\kappa_\alpha(x,y)=|x-y|^{\alpha-n}, \qquad 0<\alpha\le 2,\ n\ge 3.$$5

The source literature isolates two canonical choices of $$\kappa_\alpha(x,y)=|x-y|^{\alpha-n}, \qquad 0<\alpha\le 2,\ n\ge 3.$$6: an atomic/counting realization coming from Parseval-frame expansions, and a Gaussian path-space realization in which $$\kappa_\alpha(x,y)=|x-y|^{\alpha-n}, \qquad 0<\alpha\le 2,\ n\ge 3.$$7 and the covariance is $$\kappa_\alpha(x,y)=|x-y|^{\alpha-n}, \qquad 0<\alpha\le 2,\ n\ge 3.$$8 (Jorgensen et al., 2017).

Association also appears through factorization in de Branges–Rovnyak theory. Given a base kernel $$\kappa_\alpha(x,y)=|x-y|^{\alpha-n}, \qquad 0<\alpha\le 2,\ n\ge 3.$$9 and a contractive multiplier $A\subset \mathbb{R}^n$0, the associated kernel is

$A\subset \mathbb{R}^n$1

The recent characterization in this setting states that $A\subset \mathbb{R}^n$2 admits a complete Pick factor if and only if an auxiliary kernel $A\subset \mathbb{R}^n$3 constructed from the data satisfies that $A\subset \mathbb{R}^n$4 is itself complete Pick. An equivalent formulation uses an operator-valued holomorphic interpolation condition, and the result applies beyond complete Pick base kernels, including non-complete Pick architectures such as the Szegő kernel on the polydisk (Sau, 8 Jun 2026).

A related but distinct structural notion appears in kernel learning on the hypercube. There, Euclidean kernels on a layer $A\subset \mathbb{R}^n$5 are associated with the Johnson association scheme: their Gram matrices lie in the Bose–Mesner algebra and can be expanded in the $A\subset \mathbb{R}^n$6-basis

$A\subset \mathbb{R}^n$7

PSD reduces to linear constraints on the coefficient vector $A\subset \mathbb{R}^n$8, and the feasible set is the convex hull of $A\subset \mathbb{R}^n$9 vertex kernels. This finite spectral parametrization supports efficient MKL and universal-kernel constructions on the hypercube (Kothari et al., 2019). This suggests that, in RKHS and learning theory, associated kernels often function as symmetry-adapted coordinates on a kernel class rather than as isolated objects.

4. Support-adaptive associated kernels in nonparametric estimation

In nonparametric statistics, associated kernels are target-adaptive smoothing kernels whose support matches the support of the data. The general multivariate definition uses a pdf or pmf $\mu$0 with support $\mu$1 such that

$\mu$2

with $\mu$3 and $\mu$4 as $\mu$5. The point is to adapt both shape and support to the target $\mu$6: beta kernels respect $\mu$7, gamma kernels respect $\mu$8, and discrete associated kernels such as binomial, discrete triangular, and Dirac discrete uniform respect count or categorical supports (Somé et al., 2015).

For regression, the associated-kernel Nadaraya–Watson estimator is

$\mu$9

with product constructions for mixed supports and diagonal bandwidth matrices, or full-bandwidth correlated constructions such as the bivariate beta-Sarmanov kernel. The reported simulation evidence is nuanced. In multiple regression, matching the kernel family to the support had the dominant effect, while correlated bivariate beta kernels “did not confer practical advantages” over product beta kernels and incurred very substantial CV cost; discrete triangular kernels with small arm performed especially well for count regressions, and Epanechnikov kernels performed poorly on bounded or discrete supports because of support mismatch (Somé et al., 2015). In multivariate density estimation, however, full and Scott bandwidth matrices “generally outperform diagonal” kernels, especially for multimodal targets and when correlation is present, and a modified associated kernel can remove the first-order interior bias term by enforcing $\mu^A$0 on the interior region (Kokonendji et al., 2015). This suggests that the empirical value of correlation structure is task-dependent rather than uniform across smoothing problems.

The same support-adaptive philosophy extends to hazard estimation on $\mu^A$1. There an associated kernel is a family $\mu^A$2 whose shape depends on the estimation point $\mu^A$3 and bandwidth $\mu^A$4, with $\mu^A$5 in $\mu^A$6 as $\mu^A$7. Smoothing the increments of the Nelson–Aalen estimator yields

$\mu^A$8

Under assumptions A1–A6, the bias is $\mu^A$9 and the variance is $g_D^\alpha$00, so the general MISE-optimal rate is $g_D^\alpha$01; if $g_D^\alpha$02, the improved rate is $g_D^\alpha$03. The paper proves a CLT, oracle-type inequalities for local and global minimax bandwidth choice, and verifies all assumptions for the Gamma kernel, which is asymmetric near $g_D^\alpha$04 and therefore explicitly designed to mitigate boundary bias on $g_D^\alpha$05 (Breuil et al., 29 Sep 2025).

5. Capacity-associated and distance-induced kernels

In singular-integral potential theory, capacities can be associated with Calderón–Zygmund kernels. For

$g_D^\alpha$06

the capacity $g_D^\alpha$07 is defined for compact $g_D^\alpha$08 by requiring boundedness in $g_D^\alpha$09 of both potentials $g_D^\alpha$10 and $g_D^\alpha$11 for real distributions $g_D^\alpha$12 supported on $g_D^\alpha$13. The main theorem states that there exist constants $g_D^\alpha$14 such that

$g_D^\alpha$15

so the capacity associated with the vector kernel $g_D^\alpha$16 is quantitatively equivalent to analytic capacity. The mechanism is a symmetrization identity: the permutations $g_D^\alpha$17 are nonnegative, vanish exactly on colinear triples, and the sum $g_D^\alpha$18 is comparable to the square of the Menger curvature. The paper also emphasizes that the vectorial character is essential; single-component capacities need not enjoy the same curvature control or comparability to $g_D^\alpha$19 (Chousionis et al., 2011).

A different but closely related construction associates kernels to semimetrics of negative type. If $g_D^\alpha$20 is such a semimetric on $g_D^\alpha$21, then for any anchor $g_D^\alpha$22 the function

$g_D^\alpha$23

is PD, and the resulting MMD satisfies

$g_D^\alpha$24

For product semimetrics on $g_D^\alpha$25, HSIC with the associated kernels equals distance covariance. Characteristicness of $g_D^\alpha$26 is equivalent to strong negative type of $g_D^\alpha$27 on the relevant moment class, and on $g_D^\alpha$28 the family $g_D^\alpha$29, $g_D^\alpha$30, yields

$g_D^\alpha$31

The data explicitly note that the standard energy distance corresponds to $g_D^\alpha$32 and is only one member of a parametric family; smaller $g_D^\alpha$33 can improve sensitivity to fine-scale differences, while larger $g_D^\alpha$34 can help when differences are concentrated in means (Sejdinovic et al., 2012).

These two strands share a structural logic even though their objects differ. In both, a kernel-associated quantity becomes useful only after a positivity mechanism is identified: permutation positivity and curvature control in the Calderón–Zygmund case, and negative type in the semimetric case. This suggests that “associated kernel” constructions are often best understood as positivity-restoring transforms.

6. Terminological boundaries and adjacent usages

Not every occurrence of the phrase refers to a two-variable kernel derived from another two-variable kernel. In associative-algebra extension theory, an abstract kernel is a coupling

$g_D^\alpha$35

obtained from the action of a quotient algebra on an ideal up to inner bimultiplications. A covering $g_D^\alpha$36 and a hindrance $g_D^\alpha$37 produce the obstruction cocycle

$g_D^\alpha$38

whose class in $g_D^\alpha$39 vanishes if and only if the coupling is realizable by an extension. The paper explicitly states that associated kernels in this sense are not associated kernels from functional analysis or kernel methods (Li, 2018).

Nearby operator-theoretic usages likewise differ from the support-adaptive or RKHS meanings. In noncommutative harmonic analysis, one studies Calderón–Zygmund operators associated to matrix-valued kernels $g_D^\alpha$40 acting by left and right multiplication,

$g_D^\alpha$41

Even under standard size and smoothness assumptions, $g_D^\alpha$42-boundedness can fail for $g_D^\alpha$43 because of noncommutativity. What survives are row/column endpoint statements: weak type $g_D^\alpha$44 for perfect dyadic models and $g_D^\alpha$45 plus $g_D^\alpha$46 estimates in greater generality (Hong et al., 2012).

Other papers use “associated” in the looser sense of canonical attachment to a geometric object. Heat kernels and semigroups associated with resistance forms are attached to regular Dirichlet forms on resistance metric spaces and satisfy quantitative convergence-rate estimates under measure regularity and lower resistance assumptions (Oishi, 22 May 2026). Bergman kernels associated to positive line bundles are the Schwartz kernels of Bergman projections for high tensor powers $g_D^\alpha$47, and in the smooth Hermitian setting they satisfy the sharp off-diagonal upper bound

$g_D^\alpha$48

when $g_D^\alpha$49 (Christ, 2013). These are not alternative definitions of associated kernels as a unified concept; they are adjacent usages in which the kernel is canonically attached to a form, operator, or bundle.

Taken together, these boundaries matter as much as the constructions themselves. The phrase “associated kernel” is technically meaningful only relative to a specified ambient theory—balayage, invariance, support adaptation, negative-type geometry, extension theory, or operator attachment. Any encyclopedia treatment therefore has to preserve that contextual dependence rather than collapsing the term into a single cross-disciplinary definition.