Subinvariant kernel dynamics James Tian Abstract . We study positive definite kernels pulled back along a finite family of self-maps under a subinvariance inequality for the associated branching operator. Iteration produces an increasing kernel tower with defect kernels. Under diagonal boundedness, the tower has a smallest invariant majorant, with a canonical defect space realization and an explicit diagonal harmonic envelope governing finiteness versus blow-up. We also give probabilistic and boundary representations: a Gaussian martingale model whose quadratic variation is the defect sequence, and canonical Doob path measures with a boundary feature model for the normalized defects. 2020 Mathematics Subject Classification Primary: 46E22; secondary: 31C20, 37C30, 47A20, 60G15 1. Introduction We consider a simple but robust situation in which a positive definite kernel K:X\times X\rightarrow\mathbb{C} on a set X is repeatedly pulled back along a finite family of maps \phi_{1},\dots,\phi_{m}:X\to X . The basic object is the associated “branching” operator L acting on kernels, \left(LJ\right)\left(s,t\right):=\sum^{m}_{i=1}J\left(\phi_{i}\left(s\right),\phi_{i}\left(t\right)\right), and we assume throughout the subinvariance inequality LK\geq K in the positive-definite order. Informally, this condition says that the kernel does not lose positive mass when transported along the \phi -tree. It is reminiscent of subharmonicity and of inequalities of the form \Phi\left(A\right)\geq A for completely positive maps on operator algebras [ Pau02 ] , and it appears implicitly in a variety of settings: iterated function systems and self-similar structures, kernel-based models for non-reversible Markov dynamics, and contractive interpolation problems where a given kernel sits under a more symmetric or invariant model kernel, cf. [ LP16 , AM02 , Pau02 , PR16 ] . From a different but complementary viewpoint, operators obtained by transporting functions or observables along dynamics (Koopman and Perron-Frobenius type operators) have recently been analyzed within reproducing kernel Hilbert spaces (RKHS), where the kernel plays the role of a state space metric and transfer operators admit spectral and approximation theory in \mathcal{H}_{K} [ KSM20 ] ; related constructions produce entire families of RKHS adapted to ensemble dynamics [ DHMT24 ] . Starting from K and \left(\phi_{i}\right) (Section 2 ), the dynamics of L generates a tower of kernels K_{0}:=K,\qquad K_{n+1}:=LK_{n}\quad\left(n\geq 0\right), together with their defects D_{n}:=K_{n+1}-K_{n} . One may view K_{n} as the n -step kernel obtained by following all length- n branches in the rooted \phi -tree, while D_{n} records the new positive mass injected between levels n and n+1 . The first goal of the paper is to understand the “invariant completion” of this process: under what conditions does the increasing tower \left(K_{n}\right) converge to a limiting kernel K_{\infty} , what structure does K_{\infty} inherit from L , and how does the original kernel K sit inside this completion? Throughout, positive definiteness is understood in the classical sense of Aronszajn [ Aro50 ] (see also [ PR16 , SS16 , BTA04 ] ). A natural finiteness requirement is the diagonal boundedness condition \sup_{n\geq 0}K_{n}\left(s,s\right)<\infty\quad\text{for all }s\in X. This hypothesis is deliberately weak: we do not assume any topology or measure on X , nor any compactness or spectral input for the maps \phi_{i} . Our first main result shows that, under this assumption, the tower admits a canonical L -invariant completion. The kernels K_{n} increase pointwise and converge to a positive definite kernel K_{\infty} satisfying K_{\infty}\geq K,\qquad LK_{\infty}=K_{\infty}, and K_{\infty} is minimal among all L -invariant kernels majorizing K . Moreover, the defect tower admits an orthogonal realization: one can represent K_{\infty} on an ambient Hilbert space obtained by adjoining defect spaces level by level, so that K is recovered from K_{\infty} by a canonical compression. In particular, the subinvariance dynamics induces a concrete Radon-Nikodym type representation of K inside its own invariant completion, in a spirit close to Sz.-Nagy-Foiaş defect decompositions and model theory for contractions [ SNFBK10 ] , but carried out at the level of kernels and without assuming an operator on a pre-existing Hilbert space. The L -invariant completion also yields a natural multivariable row-isometry associated to the maps \phi_{i} , placing the construction within the scope of multivariable dilation theory [ Arv98 ] . In particular, the operator-theoretic structure of row isometries and their Wold-type decompositions [ SZ08 , Ful23 ] provides a useful organizing analogy for the orthogonal defect splitting produced here. The second theme of the paper is to understand when this completion is genuinely finite and where it breaks down (Section 3 ). The key observation is that the diagonal dynamics decouples as a purely combinatorial process on the \phi -tree. Writing u_{n}\left(s\right):=K_{n}\left(s,s\right),\qquad s\in X, one obtains a branching operator P on nonnegative functions, \left(Pu\right)\left(s\right):=\sum^{m}_{i=1}u\left(\phi_{i}\left(s\right)\right), so that u_{n+1}=Pu_{n} and hence u_{n}=P^{n}u_{0} . We show that the pointwise growth of \left(u_{n}\right) is governed by a potential-theoretic picture: the function h_{\infty}\left(s\right):=\sup_{n\geq 0}u_{n}\left(s\right) is the minimal P -harmonic majorant of u_{0}\left(s\right)=K\left(s,s\right) . This leads to a forward-invariant finiteness region X_{\mathrm{fin}}:=\left\{s\in X:\ h_{\infty}\left(s\right)<\infty\right\}, on which the invariant completion K_{\infty} is well-defined and finite on the diagonal, while outside this region the diagonal diverges monotonically along the tower. Beyond this characterization, we give usable tree-based criteria—Lyapunov-type decay conditions and complementary non-decay mechanisms based on branch counting—which are similar to classical potential theory on trees and branching Markov chains, but are formulated here for the diagonal entries of a kernel tower rather than for scalar harmonic functions [ LP16 , Woe09 , Soa94 ] . Recent work on boundary representations for harmonic and polyharmonic functions on trees [ PW19 , PW20 ] and on Martin boundaries for branching-type dynamics [ BCZ24 ] provides additional context for the role of harmonic majorants and finiteness regions in tree-indexed growth problems. A third component of the work is a probabilistic reinterpretation of the defect tower and of the compression (Section 4 ). We build a Gaussian field whose level truncations have covariance kernels K_{N} and form a martingale with orthogonal increments; the predictable quadratic variation of this martingale is the defect tower \left(D_{n}\right) . On finite sets this martingale converges precisely when the Gram matrices of K_{N} remain uniformly bounded, and the limiting covariance is K_{\infty} . At the level of the ambient Hilbert space, the compression operator becomes a literal Radon-Nikodym derivative between two Gaussian covariance structures, one with kernel K_{\infty} and one with kernel K . This point of view is in line with the well-known correspondence between Gaussian processes and RKHS [ RW06 , BCR84 , Lif12 , BTA04 ] , but here the Gaussian structure is tailored to the specific defect decomposition induced by the subinvariant dynamics. Boundary processes and jump dynamics on random trees [ Tok20 ] offer a complementary probabilistic perspective on how tree boundaries encode limiting behavior of tree-indexed evolutions. The last part (Section 5 ) is a boundary/pathspace organization of the completion. We make the implicit tree structure of the word dynamics explicit by passing to the symbolic boundary \Omega=\left\{1,\dots,m\right\}^{\mathbb{N}} and by extracting, from the minimal harmonic majorant h_{\infty} , a canonical family of Doob-transformed Markov measures \mu_{s} on \Omega . This produces a normalized “Doob” version of the branching dynamics and yields boundary cylinder expansions for the normalized kernel iterates and for the defect tower, which can then be realized as an L^{2} -boundary Gram kernel via an explicit boundary feature map. This places the present kernel subinvariance problem in the orbit of pathspace measure and boundary representation constructions for discrete branching/graph systems [ BJ22 , BJ24 , PW20 ] , while keeping the output at the level of kernels/RKHS (rather than only harmonic functions or operator-algebraic data). The overall idea is that the inequality LK\geq K , when iterated along a finite branching system of maps, forces a rich and fairly rigid structure on the invariant completion and on the diagonal growth dynamics. The resulting picture has points of contact with dilation and model theory in operator theory [ SNFBK10 , Arv98 , SZ08 , Ful23 ] , with potential theory on trees and branching processes [ LP16 , Woe09 , Soa94 , PW19 , PW20 , BCZ24 ] , and with Gaussian/RKHS methods in probability and statistics [ BTA04 , RW06 , Lif12 ] . It also aligns with the broader scope of kernel-based operator methods for dynamics, where transfer operators are studied via RKHS structure [ KSM20 , DHMT24 ] . At the same time, the setting here is combinatorial and kernel-theoretic: no measure, topology, or ambient operator algebra is required. 2. Subinvariance and invariant majorants In this section, we develop the basic subinvariance theory for the branching operator L acting on positive definite kernels. Let X be a set, K:X\times X\to\mathbb{C} a positive definite (p.d.) kernel, and \phi_{1},\dots,\phi_{m}:X\to X maps on X . We use the standard convention K_{s}\left(\cdot\right):=K\left(\cdot,s\right) , so that K\left(s,t\right)=\left\langle K_{s},K_{t}\right\rangle_{\mathcal{H}_{K}} and f\left(s\right)=\left\langle K_{s},f\right\rangle_{\mathcal{H}_{K}} , for all s\in X and f\in\mathcal{H}_{K} , where \mathcal{H}_{K} is the reproducing kernel Hilbert space (RKHS) of K . Throughout, all inner products are linear in the second variable. Let \mathcal{K}\left(X\right) denote the set of all p.d. kernels on X , and write J_{1}\geq J_{2} if J_{1}-J_{2}\in\mathcal{K}\left(X\right) . Define linear operator L acting on \mathcal{K}\left(X\right) by \left(LJ\right)\left(s,t\right):=\sum^{m}_{i=1}J\left(\phi_{i}\left(s\right),\phi_{i}\left(t\right)\right). For any J\in\mathcal{K}\left(X\right) , each pullback kernel \left(s,t\right)\mapsto J\left(\phi_{i}\left(s\right),\phi_{i}\left(t\right)\right) is p.d., hence LJ\in\mathcal{K}\left(X\right) . In particular, L is order-preserving: J_{1}\geq J_{2} implies LJ_{1}\geq LJ_{2} . Assume the subinvariance inequality (2.1) LK\geq K. Equivalently, the defect kernel D:=LK-K is positive definite, i.e., D\in\mathcal{K}\left(X\right) . For all n\in\mathbb{N}_{0}:=\left\{0,1,2,\dots\right\} , define the tower of kernels K_{n+1}:=LK_{n},\quad K_{0}:=K, and the defect iterates D_{n}:=K_{n+1}-K_{n},\quad D_{0}:=D. Finally, assume the following finiteness condition: (2.2) \sup_{n\geq 0}K_{n}\left(s,s\right)<\infty,\quad\forall s\in X. Theorem 2.1 . Under the setting above: (1) For all s,t\in X the limit (2.3) K_{\infty}\left(s,t\right):=\lim_{n\to\infty}K_{n}\left(s,t\right) exists (finite), and K_{\infty} belongs to \mathcal{K}\left(X\right) satisfying K_{\infty}\geq K,\quad LK_{\infty}=K_{\infty}. Moreover, K_{\infty} is the minimal L -invariant majorant of K . That is, if J\in\mathcal{K}\left(X\right) , J\geq K , and LJ=J , then J\geq K_{\infty} . (2) The defects satisfy D_{n}=L^{n}D , and for every N\geq 1 , (2.4) K_{N}=K+\sum^{N-1}_{n=0}D_{n}. In particular, (2.5) K_{\infty}=K+\sum^{\infty}_{n=0}D_{n}, where the series converges in the sense of Gram matrices on each finite subset of X . (3) Let \mathcal{H}_{D_{n}} be the RKHS of D_{n} , and set (2.6) \mathcal{E}:=\mathcal{H}_{K}\oplus\bigoplus_{n\geq 0}\mathcal{H}_{D_{n}}. For each s\in X define v\left(s\right):=K_{s}\oplus\bigoplus_{n\geq 0}\left(D_{n}\right)_{s}\in\mathcal{E}. Then v\left(s\right) is well-defined and K_{\infty}\left(s,t\right)=\left\langle v\left(s\right),v\left(t\right)\right\rangle_{\mathcal{E}}. Let P_{0}:\mathcal{E}\to\mathcal{H}_{K} be the coordinate projection onto the first summand, and define A:=P^{*}_{0}P_{0} (a positive contraction on \mathcal{E} ). Then K\left(s,t\right)=\left\langle v\left(s\right),Av\left(t\right)\right\rangle_{\mathcal{E}}. Thus, the subinvariance dynamics canonically produces a Radon-Nikodym compression representation of K inside the invariant completion K_{\infty} , in the sense that K_{\infty}\left(s,t\right)=\left\langle v\left(s\right),v\left(t\right)\right\rangle_{\mathcal{E}} and K\left(s,t\right)=\left\langle v\left(s\right),Av\left(t\right)\right\rangle_{\mathcal{E}} for all s,t\in X . Proof. Assumption ( 2.1 ) implies K_{1}-K_{0}=LK-K=D_{0}\geq 0. Inductively, if K_{n}\geq K_{n-1} , then K_{n+1}=LK_{n}\geq LK_{n-1}=K_{n} . Hence \left\{K_{n}\right\} is an increasing sequence. In particular, each difference D_{n}:=K_{n+1}-K_{n} is positive definite. Moreover, from K_{n+1}=LK_{n} we compute D_{n}=K_{n+1}-K_{n}=LK_{n}-LK_{n-1}=L\left(K_{n}-K_{n-1}\right)=LD_{n-1}. Thus D_{n}=L^{n}D for all n\geq 0 . Thus, for N\geq 1 , K_{N}-K=\sum^{N-1}_{n=0}\left(K_{n+1}-K_{n}\right)=\sum^{N-1}_{n=0}D_{n}, which proves the finite-level energy identity ( 2.4 ). Fix a finite set F=\left\{x_{1},\dots,x_{r}\right\}\subset X . Consider the Gram matrices G_{n}:=\left[K_{n}\left(x_{a},x_{b}\right)\right]^{r}_{a,b=1}. Each G_{n} is positive semidefinite, and G_{n+1}-G_{n}=\left[D_{n}\left(x_{a},x_{b}\right)\right] is positive semidefinite; hence G_{n} is increasing in the order on Hermitian matrices. By ( 2.2 ), the diagonal entries \left(G_{n}\right)_{aa}=K_{n}\left(x_{a},x_{a}\right) are uniformly bounded in n for each a , and therefore {\rm tr}\left(G_{n}\right)=\sum^{r}_{a=1}K_{n}\left(x_{a},x_{a}\right)\leq\sum^{r}_{a=1}\sup_{n\geq 0}K_{n}\left(x_{a},x_{a}\right)<\infty. Since G_{n}\geq 0 , one has \left\|G_{n}\right\|\leq{\rm tr}\left(G_{n}\right) , so \sup_{n}\left\|G_{n}\right\|<\infty . Thus \left\{G_{n}\right\} is increasing and uniformly bounded in operator norm, and hence converges (in operator norm) to a positive semidefinite matrix G_{\infty}:=\lim_{n\to\infty}G_{n}. In particular, for each 1\leq a,b\leq r , \left(G_{\infty}\right)_{ab}=\lim_{n\to\infty}K_{n}\left(x_{a},x_{b}\right). Since this holds for every finite F , the pointwise limit K_{\infty}\left(s,t\right):=\lim_{n\to\infty}K_{n}\left(s,t\right) defines a positive definite kernel. The monotonicity gives K_{\infty}\geq K immediately. To see LK_{\infty}=K_{\infty} , fix s,t\in X and compute \left(LK_{\infty}\right)\left(s,t\right)=\sum^{m}_{i=1}K_{\infty}\left(\phi_{i}\left(s\right),\phi_{i}\left(t\right)\right)=\sum^{m}_{i=1}\lim_{n\to\infty}K_{n}\left(\phi_{i}\left(s\right),\phi_{i}\left(t\right)\right). Since m is finite, we may pass the limit through the sum, obtaining \displaystyle\left(LK_{\infty}\right)\left(s,t\right) \displaystyle=\lim_{n\to\infty}\sum^{m}_{i=1}K_{n}\left(\phi_{i}\left(s\right),\phi_{i}\left(t\right)\right) \displaystyle=\lim_{n\to\infty}\left(LK_{n}\right)\left(s,t\right)=\lim_{n\to\infty}K_{n+1}\left(s,t\right)=K_{\infty}\left(s,t\right). For minimality: if J\geq K and LJ=J , then by positivity of L , J=L^{n}J\geq L^{n}K=K_{n}\quad\text{for all }n. Taking n\to\infty yields J\geq K_{\infty} . This completes (1). Finally, the identity ( 2.5 ) holds in the sense that for each finite F , the Gram matrices satisfy \left[K_{\infty}\left(x_{a},x_{b}\right)\right]_{a,b}=\left[K\left(x_{a},x_{b}\right)\right]_{a,b}+\sum^{\infty}_{n=0}\left[D_{n}\left(x_{a},x_{b}\right)\right]_{a,b}, with convergence in the order on Hermitian matrices because the partial sums are G_{N}-G_{0} and increase to G_{\infty}-G_{0} . This is the Gram-matrix form of (2). Define the Hilbert space \mathcal{E}:=\mathcal{H}_{K}\oplus\bigoplus_{n\geq 0}\mathcal{H}_{D_{n}}. Fix s\in X . We claim that \sum_{n\geq 0}\left\|\left(D_{n}\right)_{s}\right\|^{2}_{\mathcal{H}_{D_{n}}}<\infty, so that v\left(s\right)=K_{s}\oplus\bigoplus_{n\geq 0}\left(D_{n}\right)_{s} is a well-defined element of \mathcal{E} . Indeed, in any RKHS \mathcal{H}_{J} one has \left\|J_{s}\right\|^{2}_{\mathcal{H}_{J}}=J\left(s,s\right) . Therefore \displaystyle\sum^{N-1}_{n=0}\left\|\left(D_{n}\right)_{s}\right\|^{2}_{\mathcal{H}_{D_{n}}} \displaystyle=\sum^{N-1}_{n=0}D_{n}\left(s,s\right) \displaystyle=\sum^{N-1}_{n=0}\left(K_{n+1}\left(s,s\right)-K_{n}\left(s,s\right)\right)=K_{N}\left(s,s\right)-K\left(s,s\right). Letting N\to\infty and using ( 2.2 ) gives finiteness: \sum_{n\geq 0}\left\|\left(D_{n}\right)_{s}\right\|^{2}_{\mathcal{H}_{D_{n}}}=K_{\infty}\left(s,s\right)-K\left(s,s\right)<\infty. Hence v\left(s\right)\in\mathcal{E} for every s . Now fix s,t\in X . For each N , \left\langle K_{s}\oplus\bigoplus^{N-1}_{n=0}\left(D_{n}\right)_{s},\ K_{t}\oplus\bigoplus^{N-1}_{n=0}\left(D_{n}\right)_{t}\right\rangle_{\mathcal{E}}=K\left(s,t\right)+\sum^{N-1}_{n=0}D_{n}\left(s,t\right)=K_{N}\left(s,t\right), using the RKHS identities \left\langle K_{s},K_{t}\right\rangle_{\mathcal{H}_{K}}=K\left(s,t\right) and \left\langle\left(D_{n}\right)_{s},\left(D_{n}\right)_{t}\right\rangle_{\mathcal{H}_{D_{n}}}=D_{n}\left(s,t\right) , together with the telescoping identity. Let N\to\infty . Then K_{N}\left(s,t\right)\to K_{\infty}\left(s,t\right) , and the partial sums in \mathcal{E} converge in norm to v\left(s\right) and v\left(t\right) . Therefore the inner products converge, yielding \left\langle v\left(s\right),v\left(t\right)\right\rangle_{\mathcal{E}}=\lim_{N\to\infty}K_{N}\left(s,t\right)=K_{\infty}\left(s,t\right). This proves the feature representation in (3). Let P_{0}:\mathcal{E}\to\mathcal{H}_{K} be the coordinate projection onto the first summand, and A:=P^{*}_{0}P_{0} . Then A is a positive contraction on \mathcal{E} , and \left\langle v\left(s\right),Av\left(t\right)\right\rangle_{\mathcal{E}}=\left\langle P_{0}v\left(s\right),P_{0}v\left(t\right)\right\rangle_{\mathcal{H}_{K}}=\left\langle K_{s},K_{t}\right\rangle_{\mathcal{H}_{K}}=K\left(s,t\right), as claimed. ∎ Proposition 2.2 . Keep the setting and define K_{\infty} as above. Let \mathcal{H}_{K_{\infty}} be the RKHS of K_{\infty} . On \mathrm{span}\left\{\left(K_{\infty}\right)_{s}:s\in X\right\} define V:\mathrm{span}\left\{\left(K_{\infty}\right)_{s}:s\in X\right\}\to\mathcal{H}^{\oplus m}_{K_{\infty}},\qquad V\left(K_{\infty}\right)_{s}:=\bigoplus^{m}_{i=1}\left(K_{\infty}\right)_{\phi_{i}\left(s\right)}. Then V extends to an isometry, and writing V=\bigoplus^{m}_{i=1}V_{i} with V_{i}\in B\left(\mathcal{H}_{K_{\infty}}\right) we have \sum^{m}_{i=1}V^{*}_{i}V_{i}=I. Moreover, there exists a positive contraction A\in B\left(\mathcal{H}_{K_{\infty}}\right) such that (2.7) K\left(s,t\right)=\left\langle\left(K_{\infty}\right)_{s},A\left(K_{\infty}\right)_{t}\right\rangle_{\mathcal{H}_{K_{\infty}}}\quad\left(s,t\in X\right), and LK\geq K is equivalent to \sum^{m}_{i=1}V^{*}_{i}AV_{i}\geq A (verified on \mathrm{span}\left\{\left(K_{\infty}\right)_{s}:s\in X\right\} , hence in the form sense on \mathcal{H}_{K_{\infty}} ). Proof. From LK_{\infty}=K_{\infty} we get, for s,t\in X , \displaystyle\left\langle V\left(K_{\infty}\right)_{s},V\left(K_{\infty}\right)_{t}\right\rangle_{\mathcal{H}^{\oplus m}_{K_{\infty}}} \displaystyle=\sum^{m}_{i=1}K_{\infty}\left(\phi_{i}\left(s\right),\phi_{i}\left(t\right)\right) \displaystyle=K_{\infty}\left(s,t\right)=\left\langle\left(K_{\infty}\right)_{s},\left(K_{\infty}\right)_{t}\right\rangle_{\mathcal{H}_{K_{\infty}}}, so V is isometric on the dense span of kernel sections and extends to an isometry. Writing V=\bigoplus_{i}V_{i} then yields \left\langle f,g\right\rangle=\left\langle Vf,Vg\right\rangle=\sum_{i}\left\langle V_{i}f,V_{i}g\right\rangle=\left\langle f,\left(\sum\nolimits_{i}V^{*}_{i}V_{i}\right)g\right\rangle, hence \sum_{i}V^{*}_{i}V_{i}=I . The operator A is the compression from Theorem 2.1 , realized on \mathcal{H}_{K_{\infty}} by the identity K\left(s,t\right)=\left\langle\left(K_{\infty}\right)_{s},A\left(K_{\infty}\right)_{t}\right\rangle. Finally, for s,t\in X , \displaystyle\sum_{i}K\left(\phi_{i}\left(s\right),\phi_{i}\left(t\right)\right) \displaystyle=\sum_{i}\left\langle\left(K_{\infty}\right)_{\phi_{i}\left(s\right)},A\left(K_{\infty}\right)_{\phi_{i}\left(t\right)}\right\rangle \displaystyle=\left\langle\left(K_{\infty}\right)_{s},\left(\sum\nolimits_{i}V^{*}_{i}AV_{i}\right)\left(K_{\infty}\right)_{t}\right\rangle, so LK\geq K iff \sum_{i}V^{*}_{i}AV_{i}\geq A on \mathrm{span}\left\{\left(K_{\infty}\right)_{s}\right\} , hence in the form sense on \mathcal{H}_{K_{\infty}} . ∎ Remark 2.3 . Proposition 2.2 shows that the Radon-Nikodym compression of K may be realized internally on \mathcal{H}_{K_{\infty}} as in ( 2.7 ). In particular, the invariant completion K_{\infty} contains enough information to represent K by a compression. The auxiliary dilation space \mathcal{E} from ( 2.6 ) is nonetheless useful. First, it makes the compression operator completely explicit: on \mathcal{E} one has A=P^{*}_{0}P_{0} , where P_{0}:\mathcal{E}\to\mathcal{H}_{K} is the coordinate projection. Second, \mathcal{E} retains the defect tower as an orthogonal direct sum, so the decomposition ( 2.5 ) is realized at the Hilbert space level; this is convenient for truncations and estimates which track contributions of the individual defect kernels D_{n} . The term “Radon-Nikodym” is from the Radon-Nikodym (RN) theorem for completely positive maps on operator algebras. If \Phi,\Psi are normal completely positive maps on a C^{*} -algebra or von Neumann algebra with \Phi\leq\Psi in the usual order, then in a Stinespring dilation (\pi,H_{\Psi},V) for \Psi there exists a positive contraction 0\leq B\leq I_{H_{\Psi}} such that B\in\pi\left(\mathcal{A}\right)^{\prime} and \Phi(x)=V^{*}B\pi\left(x\right)V,\qquad x\in\mathcal{A}, see for instance [ BS86 , Arv98 , Pau02 ] for versions of this theorem. In that setting B is the RN derivative of \Phi with respect to \Psi , acting on the dilation space H_{\Psi} . The present kernel construction is a direct analogue at the level RKHS ( [ Aro50 , SS16 , Tia25 ] ). Corollary 2.4 . Let \mathcal{W}_{n}:=\left\{1,\dots,m\right\}^{n} be the set of words of length n , with \mathcal{W}_{0}:=\left\{\emptyset\right\} . For w=i_{1}\cdots i_{n}\in\mathcal{W}_{n} set \phi_{w}:=\phi_{i_{1}}\circ\cdots\circ\phi_{i_{n}},\qquad\phi_{\emptyset}:={\rm id}_{X}. Then for every J\in\mathcal{K}\left(X\right) and every n\geq 0 , \left(L^{n}J\right)\left(s,t\right)=\sum_{w\in\mathcal{W}_{n}}J\left(\phi_{w}\left(s\right),\phi_{w}\left(t\right)\right). In particular, \displaystyle K_{n}\left(s,t\right) \displaystyle=\sum_{w\in\mathcal{W}_{n}}K\left(\phi_{w}\left(s\right),\phi_{w}\left(t\right)\right), \displaystyle D_{n}\left(s,t\right) \displaystyle=\sum_{w\in\mathcal{W}_{n}}D\left(\phi_{w}\left(s\right),\phi_{w}\left(t\right)\right). Moreover, for every N\geq 1 , K_{N}\left(s,t\right)=K\left(s,t\right)+\sum^{N-1}_{n=0}\sum_{w\in\mathcal{W}_{n}}D\left(\phi_{w}\left(s\right),\phi_{w}\left(t\right)\right), and, under ( 2.2 ), K_{\infty}\left(s,t\right)=K\left(s,t\right)+\sum^{\infty}_{n=0}\sum_{w\in\mathcal{W}_{n}}D\left(\phi_{w}\left(s\right),\phi_{w}\left(t\right)\right), with convergence in the Gram-matrix sense on each finite subset of X . Proof. The formula for L^{n} follows by induction on n : the case n=0 is trivial, and the step n\to n+1 is obtained by expanding L^{n+1}J=L\left(L^{n}J\right) and regrouping the resulting sum over words of length n+1 . The identities for K_{n} and D_{n} follow from K_{n}=L^{n}K and D_{n}=L^{n}D . The finite and infinite expansions are the identities from ( 2.4 )-( 2.5 ) with these formulas inserted. ∎ We now include a concrete example illustrating the full setting above, including strict subinvariance, the finiteness hypothesis ( 2.2 ), and a nontrivial defect tower. The example is based on word dynamics on an m -ary tree, with \phi_{i} given by prefixing. The kernel is chosen so that L -invariance holds at the level of an explicit majorant while the defect decays geometrically, leading to K_{\infty} in closed form. Example 2.5 . Fix an integer m\geq 2 and let X:=\mathcal{W}:=\bigcup_{n\geq 0}\left\{1,\dots,m\right\}^{n} be the set of all finite words over \{1,\dots,m\} , including the empty word \emptyset . For w\in\mathcal{W} write \left|w\right| for the word length. Define maps \phi_{i}:X\to X by prefixing: \phi_{i}\left(w\right):=iw,\qquad 1\leq i\leq m. Define two positive definite kernels J_{0},J_{1}:X\times X\to\mathbb{C} by J_{0}\left(u,v\right):=m^{-\left|u\right|}\delta_{u,v},\qquad J_{1}\left(u,v\right):=m^{-\left(\left|u\right|+\left|v\right|\right)/2}. Then J_{0} is positive definite (diagonal with nonnegative entries), and J_{1} is positive definite (rank-one), since J_{1}\left(u,v\right)=f\left(u\right)\overline{f\left(v\right)} with f\left(w\right)=m^{-\left|w\right|/2} . Moreover, both are L -invariant: LJ_{0}=J_{0},\qquad LJ_{1}=J_{1}. Indeed, for s,t\in X , \left(LJ_{0}\right)\left(s,t\right)=\sum^{m}_{i=1}m^{-\left|is\right|}\delta_{is,it}=\sum^{m}_{i=1}m^{-\left(\left|s\right|+1\right)}\delta_{s,t}=J_{0}\left(s,t\right), and \left(LJ_{1}\right)\left(s,t\right)=\sum^{m}_{i=1}m^{-\left(\left|is\right|+\left|it\right|\right)/2}=\sum^{m}_{i=1}m^{-\left(\left|s\right|+\left|t\right|\right)/2-1}=J_{1}\left(s,t\right). Fix parameters 00 . Define a diagonal positive definite kernel E by E\left(u,v\right):=cr^{\left|u\right|}m^{-\left|u\right|}\delta_{u,v}. Then LE=rE , since for s\in X , \left(LE\right)\left(s,s\right)=\sum^{m}_{i=1}E\left(is,is\right)=\sum^{m}_{i=1}cr^{\left|is\right|}m^{-\left|is\right|}=rcr^{\left|s\right|}m^{-\left|s\right|}=rE\left(s,s\right). Now set J:=J_{0}+\eta J_{1},\qquad K:=J-E. Since J_{0}-E is diagonal with nonnegative entries (because 00 may be viewed as a controlled “correlation” perturbation of the diagonal model which preserves the subinvariance dynamics and the finiteness condition ( 2.2 ). 3. Diagonal dynamics The finiteness hypothesis ( 2.2 ) and the existence of the invariant completion are governed by the diagonal growth of the tower. In this section, we isolate the diagonal data u_{n}\left(s\right)=K_{n}\left(s,s\right) and show that it evolves under the induced branching operator P on functions, u_{n+1}=Pu_{n} . This reduces diagonal boundedness to a purely combinatorial problem on the rooted \phi -tree. We identify the envelope h_{\infty}\left(s\right)=\sup_{n\geq 0}u_{n}\left(s\right) as the minimal P -harmonic majorant of u_{0} , introduce the forward-invariant finiteness region X_{\mathrm{fin}} , and derive usable criteria, via Lyapunov superharmonic functions and branch counting estimates, for deciding whether h_{\infty}\left(s\right) is finite or infinite. We begin with u_{0}\left(s\right):=K\left(s,s\right)\in\left[0,\infty\right], and define the positive operator P on nonnegative functions u:X\to\left[0,\infty\right] by \left(Pu\right)\left(s\right):=\sum^{m}_{i=1}u\left(\phi_{i}\left(s\right)\right). Recall that, for a word w=i_{1}\cdots i_{n}\in\mathcal{W} , we set \phi_{w}:=\phi_{i_{1}}\circ\cdots\circ\phi_{i_{n}}, and we have K_{n+1}=LK_{n} , so K_{n}=L^{n}K . Set u_{n}\left(s\right):=K_{n}\left(s,s\right). Lemma 3.1 . For every n\geq 0 and s\in X , u_{n+1}\left(s\right)=\left(Pu_{n}\right)\left(s\right) , hence u_{n}=P^{n}u_{0} . Equivalently, (3.1) u_{n}\left(s\right)=\sum_{\left|w\right|=n}u_{0}\left(\phi_{w}\left(s\right)\right). Proof. By definition, \displaystyle u_{n+1}\left(s\right) \displaystyle=K_{n+1}\left(s,s\right)=\left(LK_{n}\right)\left(s,s\right) \displaystyle=\sum^{m}_{i=1}K_{n}\left(\phi_{i}\left(s\right),\phi_{i}\left(s\right)\right)=\sum^{m}_{i=1}u_{n}\left(\phi_{i}\left(s\right)\right)=\left(Pu_{n}\right)\left(s\right). Iterating gives u_{n}=P^{n}u_{0} , and expanding P^{n} yields the \phi -tree formula ( 3.1 ). ∎ A first consequence is that ( 2.2 ) is entirely a \phi -driven growth condition (Theorem 3.2 ), i.e., it is equivalent to (3.2) \sup_{n\geq 0}\left(P^{n}u_{0}\right)\left(s\right)<\infty,\quad\forall s\in X. This already makes the maps \phi_{i} visible: it is not merely “about L ” in the abstract, but about the combinatorics of the rooted \phi -tree w\mapsto\phi_{w}\left(s\right) and how the diagonal weight u_{0}\left(s\right)=K\left(s,s\right) decays along it. Theorem 3.2 . Let h_{\infty}\left(s\right):=\sup_{n\geq 0}u_{n}\left(s\right)=\sup_{n\geq 0}\left(P^{n}u_{0}\right)\left(s\right)\in\left[0,\infty\right]. Then the following hold: (1) h_{\infty} is the minimal P -harmonic majorant of u_{0} in the following sense: if h:X\to\left[0,\infty\right] satisfies h\geq u_{0},\qquad Ph=h, then h\geq h_{\infty} . (2) Let X_{\mathrm{fin}}:=\left\{s\in X:\ h_{\infty}\left(s\right)<\infty\right\},\qquad X_{\mathrm{inf}}:=X\setminus X_{\mathrm{fin}}. Then X_{\mathrm{fin}} is forward invariant for the maps \phi_{i} : if s\in X_{\mathrm{fin}} , then \phi_{i}\left(s\right)\in X_{\mathrm{fin}} for every i . Equivalently, if \phi_{i}\left(s\right)\in X_{\mathrm{inf}} for some i , then s\in X_{\mathrm{inf}} . (3) The finiteness hypothesis ( 2.2 ) holds if and only if X_{\mathrm{fin}}=X . Proof. Fix h with h\geq u_{0} and Ph=h . Then we have P^{n}h=h for all n\geq 0 . Moreover, since P is positive, h\geq u_{0} implies P^{n}h\geq P^{n}u_{0}=u_{n} for all n . Hence h\geq u_{n} for every n , and therefore h\geq\sup_{n\geq 0}u_{n}=h_{\infty}. This proves the minimality statement in (1). Note that, since LK\geq K , the kernels K_{n+1}-K_{n} are positive definite, hence u_{n+1}\left(s\right)-u_{n}\left(s\right)=\left(K_{n+1}-K_{n}\right)\left(s,s\right)\geq 0 for every s . Thus u_{n}\left(s\right)\uparrow h_{\infty}\left(s\right) pointwise, and because m is finite we may pass to the limit in the defining sum for P : for every s\in X , \displaystyle\left(Ph_{\infty}\right)\left(s\right) \displaystyle=\sum^{m}_{i=1}h_{\infty}\left(\phi_{i}\left(s\right)\right)=\sum^{m}_{i=1}\lim_{n\to\infty}u_{n}\left(\phi_{i}\left(s\right)\right) \displaystyle=\lim_{n\to\infty}\sum^{m}_{i=1}u_{n}\left(\phi_{i}\left(s\right)\right)=\lim_{n\to\infty}u_{n+1}\left(s\right)=h_{\infty}\left(s\right). In particular, h_{\infty} itself is P -harmonic. For (2), assume h_{\infty}\left(s\right)<\infty . Then u_{n+1}\left(s\right)=\sum_{i}u_{n}\left(\phi_{i}\left(s\right)\right)\leq h_{\infty}\left(s\right) for all n . Since each term is nonnegative, for each fixed i we have u_{n}\left(\phi_{i}\left(s\right)\right)\leq u_{n+1}\left(s\right)\leq h_{\infty}\left(s\right) for all n , hence h_{\infty}\left(\phi_{i}\left(s\right)\right)<\infty . This is the forward invariance. Finally, (3) is immediate from the definition: ( 2.2 ) is the pointwise finiteness of \sup_{n}u_{n}\left(s\right) , i.e., X_{\mathrm{fin}}=X . ∎ Remark 3.3 . The diagonal growth dichotomy governs where the invariant completion from Theorem 2.1 is genuinely finite. If s\in X_{\mathrm{inf}} , then u_{n}\left(s\right)=K_{n}\left(s,s\right)\to\infty , so the diagonal of the tower diverges at s and the finiteness hypothesis ( 2.2 ) fails there. On the other hand, on X_{\mathrm{fin}} one has \sup_{n}K_{n}\left(s,s\right)<\infty , and the Gram-matrix argument in Theorem 2.1 (1) applies on each finite subset F\subset X_{\mathrm{fin}} , giving a well-defined limit kernel K_{\infty} on X_{\mathrm{fin}}\times X_{\mathrm{fin}} with K_{\infty}\geq K and LK_{\infty}=K_{\infty} restricted to X_{\mathrm{fin}} . Thus, Section 2 may be viewed as operating on the forward-invariant region X_{\mathrm{fin}} , while the present section provides dynamical criteria to determine when X_{\mathrm{fin}}=X . A useful conceptual point is that this theorem is already a sharp “potential-theoretic” characterization driven by the \phi -tree: the diagonal growth is finite exactly when the diagonal data u_{0}\left(s\right)=K\left(s,s\right) lies under a harmonic function for the branching operator P . The abstract characterization above becomes usable once one has concrete sufficient conditions phrased directly in terms of the maps \phi_{i} . The following criterion (Corollary 3.4 ) makes the \phi_{i} explicit: it asks for a “Lyapunov function” r that does not grow (on average) along the \phi -branches. Corollary 3.4 . Assume there exists a function r:X\to\left(0,\infty\right) and a constant C<\infty such that u_{0}\leq Cr,\qquad Pr\leq r. Then h_{\infty}\left(s\right)<\infty for all s , hence ( 3.2 ) holds. Proof. We show by induction that u_{n}\leq Cr for all n . For n=0 this is u_{0}\leq Cr . If u_{n}\leq Cr , then u_{n+1}=Pu_{n}\leq CPr\leq Cr. Thus u_{n}\left(s\right)\leq Cr\left(s\right) for all n , so h_{\infty}\left(s\right)\leq Cr\left(s\right)<\infty . ∎ The next result (Corollary 3.5 ) is a complementary “non-decay” mechanism: if there is a region where the diagonal kernel mass u_{0}\left(s\right)=K\left(s,s\right) is bounded below, and the \phi -tree hits that region along exponentially many branches, then the diagonal growth must blow up. Corollary 3.5 . Fix s\in X . Suppose there exists a subset Y\subset X , constants \varepsilon>0 , \rho>1 , and an infinite sequence n_{k}\to\infty such that u_{0}\left(y\right)\geq\varepsilon\ \text{for all }y\in Y,\qquad\#\left\{w:\ \left|w\right|=n_{k},\ \phi_{w}\left(s\right)\in Y\right\}\ \geq\ \rho^{n_{k}}\ \text{for all }k. Then h_{\infty}\left(s\right)=\infty , so s\in X_{\mathrm{inf}} . Proof. By the \phi -tree expansion, \displaystyle u_{n_{k}}\left(s\right) \displaystyle=\sum_{\left|w\right|=n_{k}}u_{0}\left(\phi_{w}\left(s\right)\right) \displaystyle\geq\sum_{\begin{subarray}{c}\left|w\right|=n_{k}\\ \phi_{w}\left(s\right)\in Y\end{subarray}}u_{0}\left(\phi_{w}\left(s\right)\right)\geq\varepsilon\ \#\left\{w:\left|w\right|=n_{k},\ \phi_{w}\left(s\right)\in Y\right\}\geq\varepsilon\rho^{n_{k}}. Hence u_{n_{k}}\left(s\right)\to\infty , so h_{\infty}\left(s\right)=\infty . ∎ We now characterize the diagonal growth entirely in terms of \phi -tree level-set counts. Theorem 3.6 . Fix s\in X . For each \theta>0 and n\geq 0 define N_{n}\left(s,\theta\right):=\#\left\{w:\left|w\right|=n,u_{0}\left(\phi_{w}\left(s\right)\right)\geq\theta\right\}, where u_{0}\left(s\right)=K\left(s,s\right) and \phi_{w}=\phi_{i_{1}}\circ\cdots\circ\phi_{i_{n}} for a word w=i_{1}\cdots i_{n}\in\mathcal{W} . Then: (1) For every n\geq 0 , (3.3) u_{n}\left(s\right)=\int^{\infty}_{0}N_{n}\left(s,\theta\right)d\theta. (2) Consequently, (3.4) h_{\infty}\left(s\right)=\sup_{n\geq 0}u_{n}\left(s\right)\leq\int^{\infty}_{0}\left(\sup_{n\geq 0}N_{n}\left(s,\theta\right)\right)d\theta. (3) In particular: (a) If there exists \theta_{0}>0 such that \sup_{n\geq 0}N_{n}\left(s,\theta_{0}\right)=\infty , then h_{\infty}\left(s\right)=\infty , hence s\in X_{\mathrm{inf}} . (b) If (3.5) \int^{\infty}_{0}\left(\sup_{n\geq 0}N_{n}\left(s,\theta\right)\right)d\theta<\infty, then h_{\infty}\left(s\right)<\infty , hence s\in X_{\mathrm{fin}} and so ( 3.2 ) holds at s . Proof. Fix s\in X and n\geq 0 . By Lemma 3.1 , u_{n}\left(s\right)=\sum_{\left|w\right|=n}u_{0}\left(\phi_{w}\left(s\right)\right). For each a\in\left[0,\infty\right) one has the identity a=\int^{\infty}_{0}\mathbf{1}_{\left\{a\geq\theta\right\}}d\theta. Applying this with a=u_{0}\left(\phi_{w}\left(s\right)\right) and summing over \left|w\right|=n gives \displaystyle u_{n}\left(s\right) \displaystyle=\sum_{\left|w\right|=n}\int^{\infty}_{0}\mathbf{1}_{\left\{u_{0}\left(\phi_{w}\left(s\right)\right)\geq\theta\right\}}d\theta \displaystyle=\int^{\infty}_{0}\left(\sum_{\left|w\right|=n}\mathbf{1}_{\left\{u_{0}\left(\phi_{w}\left(s\right)\right)\geq\theta\right\}}\right)d\theta=\int^{\infty}_{0}N_{n}\left(s,\theta\right)d\theta, where we interchange sum and integral by nonnegativity. This proves ( 3.3 ). For ( 3.4 ), observe that for each n , \int^{\infty}_{0}N_{n}\left(s,\theta\right)d\theta\leq\int^{\infty}_{0}\left(\sup_{k\geq 0}N_{k}\left(s,\theta\right)\right)d\theta. Taking the supremum over n yields ( 3.4 ). For (3), if \sup_{n}N_{n}\left(s,\theta_{0}\right)=\infty , then for every n , using that \theta\mapsto N_{n}\left(s,\theta\right) is non-increasing, \displaystyle u_{n}\left(s\right) \displaystyle=\int^{\infty}_{0}N_{n}\left(s,\theta\right)d\theta \displaystyle\geq\int^{\theta_{0}}_{0}N_{n}\left(s,\theta\right)d\theta\geq\int^{\theta_{0}}_{0}N_{n}\left(s,\theta_{0}\right)d\theta=\theta_{0}N_{n}\left(s,\theta_{0}\right), so \sup_{n}u_{n}\left(s\right)=\infty , i.e. h_{\infty}\left(s\right)=\infty . The finiteness condition in ( 3.5 ) implies h_{\infty}\left(s\right)<\infty by ( 3.4 ), hence s\in X_{\mathrm{fin}} . ∎ Remark 3.7 . Corollaries 3.4 and 3.5 are obtained by estimating N_{n}\left(s,\theta\right) from below on a fixed level set \left\{u_{0}\geq\theta\right\} (the non-decay mechanism) and from above by dominating u_{0} by a P -superharmonic Lyapunov function (the decay mechanism), which in turn forces uniform control of the integrand in ( 3.4 ). The next estimate explains why the diagonal criteria from this section are already useful for controlling the full invariant completion: once the diagonal tail h_{\infty}-u_{N} is small, the entire kernel tail K_{\infty}-K_{N} is small uniformly on finite sets. Theorem 3.8 . Fix N\geq 0 . On X_{\mathrm{fin}} the limit kernel K_{\infty} from Theorem 2.1 is well-defined, and the tail kernel R_{N}\left(s,t\right):=K_{\infty}\left(s,t\right)-K_{N}\left(s,t\right),\qquad s,t\in X_{\mathrm{fin}}, is positive definite. Moreover, for all s,t\in X_{\mathrm{fin}} one has (3.6) \left|K_{\infty}\left(s,t\right)-K_{N}\left(s,t\right)\right|^{2}\leq\left(K_{\infty}\left(s,s\right)-K_{N}\left(s,s\right)\right)\left(K_{\infty}\left(t,t\right)-K_{N}\left(t,t\right)\right). Equivalently, writing u_{N}\left(s\right)=K_{N}\left(s,s\right) and h_{\infty}\left(s\right)=K_{\infty}\left(s,s\right) on X_{\mathrm{fin}} , (3.7) \left|K_{\infty}\left(s,t\right)-K_{N}\left(s,t\right)\right|\leq\sqrt{\left(h_{\infty}\left(s\right)-u_{N}\left(s\right)\right)\left(h_{\infty}\left(t\right)-u_{N}\left(t\right)\right)}. Proof. Fix a finite subset F=\left\{x_{1},\dots,x_{r}\right\}\subset X_{\mathrm{fin}} . Since x_{a}\in X_{\mathrm{fin}} , the diagonal sequence K_{n}\left(x_{a},x_{a}\right) is bounded in n , hence the Gram-matrix argument from Theorem 2.1 (1) applies on F and yields the limit matrix G_{\infty}:=\lim_{n\to\infty}\left[K_{n}\left(x_{a},x_{b}\right)\right]_{a,b}. In particular, K_{\infty} exists on F\times F and the Gram matrices satisfy \left[K_{N}\left(x_{a},x_{b}\right)\right]_{a,b}\leq\left[K_{\infty}\left(x_{a},x_{b}\right)\right]_{a,b} in the order on Hermitian matrices (because \left\{K_{n}\right\} is increasing). Therefore, for each N , the difference matrix \left[R_{N}\left(x_{a},x_{b}\right)\right]_{a,b}=\left[K_{\infty}\left(x_{a},x_{b}\right)-K_{N}\left(x_{a},x_{b}\right)\right]_{a,b} is positive semidefinite. Since this holds for every finite F\subset X_{\mathrm{fin}} , it follows that R_{N}\in\mathcal{K}\left(X_{\mathrm{fin}}\right) . Let \mathcal{H}_{R_{N}} be the RKHS of R_{N} . For s,t\in X_{\mathrm{fin}} , the reproducing property gives R_{N}\left(s,t\right)=\left\langle\left(R_{N}\right)_{s},\left(R_{N}\right)_{t}\right\rangle_{\mathcal{H}_{R_{N}}}. Hence, by Cauchy-Schwarz, \left|R_{N}\left(s,t\right)\right|^{2}\leq\left\|\left(R_{N}\right)_{s}\right\|^{2}_{\mathcal{H}_{R_{N}}}\left\|\left(R_{N}\right)_{t}\right\|^{2}_{\mathcal{H}_{R_{N}}}=R_{N}\left(s,s\right)R_{N}\left(t,t\right), using \left\|J_{s}\right\|^{2}_{\mathcal{H}_{J}}=J\left(s,s\right) for any p.d. kernel J . Substituting R_{N}=K_{\infty}-K_{N} yields ( 3.6 ), and ( 3.7 ) is just the same inequality with R_{N}\left(s,s\right)=h_{\infty}\left(s\right)-u_{N}\left(s\right) . ∎ A convenient way to use Theorem 3.8 is to bound the diagonal tail h_{\infty}-u_{N} via the defect tower. Corollary 3.9 . Let d_{n}\left(s\right):=D_{n}\left(s,s\right) and d_{0}\left(s\right)=D\left(s,s\right) . Then on X_{\mathrm{fin}} , for every N\geq 0 one has (3.8) h_{\infty}\left(s\right)-u_{N}\left(s\right)=\sum_{n\geq N}d_{n}\left(s\right). In particular, suppose there exist a function r:X\to\left(0,\infty\right) , constants C<\infty and \beta\in\left(0,1\right) such that d_{0}\leq Cr,\qquad Pr\leq\beta r. Then for every N\geq 0 and s,t\in X_{\mathrm{fin}} , (3.9) \left|K_{\infty}\left(s,t\right)-K_{N}\left(s,t\right)\right|\leq\frac{C}{1-\beta}\beta^{N}\sqrt{r\left(s\right)r\left(t\right)}. Proof. Since K_{\infty}=K+\sum_{n\geq 0}D_{n} in Gram-matrix sense on X_{\mathrm{fin}} (see Theorem 2.1 ), taking diagonals gives h_{\infty}\left(s\right)-u_{N}\left(s\right)=K_{\infty}\left(s,s\right)-K_{N}\left(s,s\right)=\sum_{n\geq N}D_{n}\left(s,s\right)=\sum_{n\geq N}d_{n}\left(s\right), which is ( 3.8 ). Next, note that d_{n}\left(s\right)=D_{n}\left(s,s\right)=\left(L^{n}D\right)\left(s,s\right) , so by the same diagonal computation as in Lemma 3.1 one has d_{n+1}=Pd_{n},\qquad d_{n}=P^{n}d_{0}. Hence d_{0}\leq Cr and Pr\leq\beta r imply inductively that d_{n}=P^{n}d_{0}\leq CP^{n}r\leq C\beta^{n}r. Therefore h_{\infty}\left(s\right)-u_{N}\left(s\right)=\sum_{n\geq N}d_{n}\left(s\right)\leq\sum_{n\geq N}C\beta^{n}r\left(s\right)=\frac{C}{1-\beta}\beta^{N}r\left(s\right). Combining this diagonal tail estimate with Theorem 3.8 gives ( 3.9 ). ∎ 4. Gaussian defect martingales Theorem 2.1 gives a canonical splitting of the invariant completion K_{\infty} into the initial kernel K and the defect kernels D_{n} . In this section we turn that splitting into a probabilistic object: a Gaussian process whose level truncations form a martingale, and whose predictable quadratic variation is the defect tower. This gives a concrete Radon-Nikodym interpretation of the compression operator A . Theorem 4.1 . Assume the setting of Section 2 , and define \mathcal{E}:=\mathcal{H}_{K}\oplus\bigoplus_{n\geq 0}\mathcal{H}_{D_{n}},\qquad v\left(s\right):=K_{s}\oplus\bigoplus_{n\geq 0}\left(D_{n}\right)_{s}\in\mathcal{E},\qquad A:=P^{*}_{0}P_{0}\in B\left(\mathcal{E}\right) as in Theorem 2.1 (3). Let \left(G_{0},G_{1},G_{2},\dots\right) be a sequence of independent centered complex Gaussian Hilbert-space vectors with G_{0}\in\mathcal{H}_{K},\qquad G_{n+1}\in\mathcal{H}_{D_{n}}\quad\left(n\geq 0\right), each normalized so that for all f,g in the corresponding Hilbert space, \mathbb{E}\left[\overline{\left\langle f,G\right\rangle}\left\langle g,G\right\rangle\right]=\left\langle f,g\right\rangle. For N\geq 0 define the truncated random field X_{N}\left(s\right):=\left\langle K_{s},G_{0}\right\rangle_{\mathcal{H}_{K}}+\sum^{N-1}_{n=0}\left\langle\left(D_{n}\right)_{s},G_{n+1}\right\rangle_{\mathcal{H}_{D_{n}}},\qquad s\in X, with the convention that the sum is 0 when N=0 . Let \mathcal{F}_{N} be the \sigma -algebra generated by \left(G_{0},G_{1},\dots,G_{N}\right) . Then the following hold. (1) For every N\geq 0 and all s,t\in X , \mathbb{E}\left[\overline{X_{N}\left(s\right)}X_{N}\left(t\right)\right]=K_{N}\left(s,t\right). In particular, each finite-dimensional distribution of \left(X_{N}\left(s\right)\right)_{s\in X} is centered complex Gaussian with covariance kernel K_{N} . (2) For each fixed s\in X , the sequence \left(X_{N}\left(s\right)\right)_{N\geq 0} is an \left(\mathcal{F}_{N}\right) -martingale in L^{2} , with orthogonal increments X_{N+1}\left(s\right)-X_{N}\left(s\right)=\left\langle\left(D_{N}\right)_{s},G_{N+1}\right\rangle_{\mathcal{H}_{D_{N}}}. Moreover, for all s,t\in X , \mathbb{E}\left[\overline{\left(X_{N+1}\left(s\right)-X_{N}\left(s\right)\right)}\left(X_{N+1}\left(t\right)-X_{N}\left(t\right)\right)\ \middle|\ \mathcal{F}_{N}\right]=D_{N}\left(s,t\right), thus the predictable quadratic variation of the martingale is the defect tower. (3) Fix a finite subset F=\left\{x_{1},\dots,x_{r}\right\}\subset X . Then \left(X_{N}\left(x_{a}\right)\right)^{r}_{a=1} converges in L^{2} (and almost surely) as N\to\infty if and only if the Gram matrices \left[K_{N}\left(x_{a},x_{b}\right)\right]_{a,b} are uniformly bounded, equivalently if and only if the limit Gram matrix G_{\infty}:=\lim_{N\to\infty}\left[K_{N}\left(x_{a},x_{b}\right)\right]_{a,b} exists as in the Gram-matrix argument of Theorem 2.1 (1). In that case the limit X_{\infty}\left(x_{a}\right):=\lim_{N\to\infty}X_{N}\left(x_{a}\right) is centered complex Gaussian and satisfies \mathbb{E}\left[\overline{X_{\infty}\left(x_{a}\right)}X_{\infty}\left(x_{b}\right)\right]=K_{\infty}\left(x_{a},x_{b}\right)\qquad\left(1\leq a,b\leq r\right). (4) Let G:=G_{0}\oplus\bigoplus_{n\geq 0}G_{n+1}\in\mathcal{E} , and define two random fields on X by Z\left(s\right):=\left\langle v\left(s\right),G\right\rangle_{\mathcal{E}},\qquad Y\left(s\right):=\left\langle v\left(s\right),A^{1/2}G\right\rangle_{\mathcal{E}}. Then Z is centered complex Gaussian with covariance K_{\infty} , while Y is centered complex Gaussian with covariance K , namely \mathbb{E}\left[\overline{Z\left(s\right)}Z\left(t\right)\right]=K_{\infty}\left(s,t\right),\qquad\mathbb{E}\left[\overline{Y\left(s\right)}Y\left(t\right)\right]=K\left(s,t\right). Thus, A is the Radon-Nikodym compression operator in the sense of a Gaussian covariance contraction. Proof. (1) Fix N and s,t\in X . Independence of the Gaussian vectors across levels gives \mathbb{E}\left[\overline{X_{N}\left(s\right)}X_{N}\left(t\right)\right]=\mathbb{E}\left[\overline{\left\langle K_{s},G_{0}\right\rangle}\left\langle K_{t},G_{0}\right\rangle\right]+\sum^{N-1}_{n=0}\mathbb{E}\left[\overline{\left\langle\left(D_{n}\right)_{s},G_{n+1}\right\rangle}\left\langle\left(D_{n}\right)_{t},G_{n+1}\right\rangle\right]. By the normalization of each Gaussian vector and the RKHS identity \left\langle J_{s},J_{t}\right\rangle_{\mathcal{H}_{J}}=J\left(s,t\right) , the first term equals K\left(s,t\right) and the n th term in the sum equals D_{n}\left(s,t\right) . Hence \mathbb{E}\left[\overline{X_{N}\left(s\right)}X_{N}\left(t\right)\right]=K\left(s,t\right)+\sum^{N-1}_{n=0}D_{n}\left(s,t\right)=K_{N}\left(s,t\right) by ( 2.4 ). This proves (1). (2) Fix s\in X . Since X_{N}\left(s\right) depends only on \left(G_{0},\dots,G_{N}\right) , it is \mathcal{F}_{N} -measurable. Moreover, X_{N+1}\left(s\right)=X_{N}\left(s\right)+\left\langle\left(D_{N}\right)_{s},G_{N+1}\right\rangle. Because G_{N+1} is independent of \mathcal{F}_{N} and centered, \mathbb{E}\left[X_{N+1}\left(s\right)\ \middle|\ \mathcal{F}_{N}\right]=X_{N}\left(s\right)+\mathbb{E}\left[\left\langle\left(D_{N}\right)_{s},G_{N+1}\right\rangle\right]=X_{N}\left(s\right), so \left(X_{N}\left(s\right)\right) is a martingale. Orthogonality of increments follows from independence across levels. For the conditional covariance identity, again using independence and the normalization, \mathbb{E}\left[\overline{\left\langle\left(D_{N}\right)_{s},G_{N+1}\right\rangle}\left\langle\left(D_{N}\right)_{t},G_{N+1}\right\rangle\ \middle|\ \mathcal{F}_{N}\right]=\left\langle\left(D_{N}\right)_{s},\left(D_{N}\right)_{t}\right\rangle_{\mathcal{H}_{D_{N}}}=D_{N}\left(s,t\right), which is the claimed predictable quadratic variation identity. (3) Fix F=\left\{x_{1},\dots,x_{r}\right\} . By (1), the covariance matrix of the centered Gaussian vector \left(X_{N}\left(x_{a}\right)\right)^{r}_{a=1} is the Gram matrix G_{N}=\left[K_{N}\left(x_{a},x_{b}\right)\right]_{a,b} . The sequence converges in L^{2} (equivalently, is Cauchy in L^{2} ) if and only if \sup_{N}\left\|G_{N}\right\|<\infty , which is equivalent to the existence of the limit Gram matrix G_{\infty} as in the Gram-matrix argument of Theorem 2.1 (1). In that case the limit is necessarily centered Gaussian with covariance G_{\infty} , hence corresponds to K_{\infty} on F\times F . (4) The random vector G\in\mathcal{E} is centered Gaussian with covariance equal to the identity on \mathcal{E} in the same sense as above. Therefore, for Z\left(s\right)=\left\langle v\left(s\right),G\right\rangle , \mathbb{E}\left[\overline{Z\left(s\right)}Z\left(t\right)\right]=\left\langle v\left(s\right),v\left(t\right)\right\rangle_{\mathcal{E}}=K_{\infty}\left(s,t\right) by Theorem 2.1 (3). Similarly, since A^{1/2} is a contraction, \mathbb{E}\left[\overline{Y\left(s\right)}Y\left(t\right)\right]=\left\langle v\left(s\right),Av\left(t\right)\right\rangle_{\mathcal{E}}=K\left(s,t\right), again by Theorem 2.1 (3). This completes the proof. ∎ Remark 4.2 . The theorem is a probabilistic interpretation of the defect tower: X_{N} is the level- N partial energy, and D_{N} is the conditional covariance of the N th increment. In particular, on each finite subset F\subset X the convergence of the invariant completion K_{\infty} is equivalent to the L^{2} -boundedness of this martingale, so the finiteness hypothesis ( 2.2 ) is a martingale boundedness condition expressed on diagonals. 5. Boundary model and Doob transform We now make explicit the tree/pathspace picture which has been implicit in our discussion, and we obtain the canonical path measures attached to the minimal harmonic majorant h_{\infty} from Section 3 . The point is that the diagonal harmonic object h_{\infty} produces a natural family of Markov measures on the boundary \Omega , and these measures organize the iterates of L (after a canonical normalization) as boundary averages. Throughout this section we assume the standing hypotheses of Sections 2 – 4 , and we write h:=h_{\infty} for the minimal P -harmonic majorant of u_{0}\left(s\right)=K\left(s,s\right) (constructed in Section 3 ). We recall that h is P -harmonic, meaning h\left(s\right)=\sum^{m}_{i=1}h\left(\varphi_{i}\left(s\right)\right), and that h\left(s\right)\in\left(0,\infty\right) precisely on the set X_{\mathrm{fin}} where the diagonal growth is finite. In what follows we work on X^{+}_{\mathrm{fin}}:=\left\{s\in X_{\mathrm{fin}}:h\left(s\right)>0\right\}, where K_{\infty} is defined by the finite-set Gram-matrix limits. Let \Omega:=\left\{1,\dots,m\right\}^{\mathbb{N}} be the pathspace, equipped with its product \sigma -algebra \mathcal{F} generated by cylinder sets. For a finite word w=i_{1}\cdots i_{n}\in\left\{1,\dots,m\right\}^{n} we write \left[w\right]:=\left\{\omega\in\Omega:\omega_{1}\cdots\omega_{n}=w\right\}. We also write \omega|n=\omega_{1}\cdots\omega_{n} for the length- n prefix. For s\in X and w=i_{1}\cdots i_{n} we keep the usual composition notation \varphi_{w}=\varphi_{i_{1}}\circ\cdots\circ\varphi_{i_{n}},\qquad\varphi_{\emptyset}=\mathrm{id}_{X}, and we also define the reversed composition \check{\varphi}_{w}:=\varphi_{i_{n}}\circ\cdots\circ\varphi_{i_{1}},\qquad\check{\varphi}_{\emptyset}=\mathrm{id}_{X}. We then define the branch points along \omega by s_{n}\left(\omega\right):=\check{\varphi}_{\omega|n}\left(s\right),\qquad n\geq 0. The harmonic function h defines canonical transition probabilities on X_{\mathrm{fin}} . For s\in X^{+}_{\mathrm{fin}} , set p_{i}\left(s\right):=\frac{h\left(\varphi_{i}\left(s\right)\right)}{h\left(s\right)},\qquad i=1,\dots,m. Then p_{i}\left(s\right)\geq 0 and \sum^{m}_{i=1}p_{i}\left(s\right)=1 for every s\in X^{+}_{\mathrm{fin}} , by P -harmonicity of h . Proposition 5.1 . For each s\in X^{+}_{\mathrm{fin}} there is a unique probability measure \mu_{s} on \left(\Omega,\mathcal{F}\right) such that for every word w=i_{1}\cdots i_{n} , (5.1) \mu_{s}\left(\left[w\right]\right)=\frac{h\left(\check{\varphi}_{w}\left(s\right)\right)}{h\left(s\right)}=\prod^{n}_{k=1}p_{i_{k}}\left(\check{\varphi}_{i_{1}\cdots i_{k-1}}\left(s\right)\right). Moreover, if \mathcal{F}_{n} denotes the \sigma -algebra generated by cylinders of length n , then under \mu_{s} the process \left(s_{n}\left(\omega\right)\right)_{n\geq 0} is a time-homogeneous Markov chain with transitions (5.2) \mu_{s}\left(s_{n+1}=\varphi_{i}\left(x\right)\mid s_{n}=x\right)=p_{i}\left(x\right),\qquad x\in X^{+}_{\mathrm{fin}}. Equivalently, for every bounded function f on X^{+}_{\mathrm{fin}} and every n\geq 0 , (5.3) \mathbb{E}_{\mu_{s}}\left[f\left(s_{n+1}\right)\mid\mathcal{F}_{n}\right]=\sum^{m}_{i=1}p_{i}\left(s_{n}\right)f\left(\varphi_{i}\left(s_{n}\right)\right). Proof. Define a set function on cylinders by \mu_{s}\left(\left[w\right]\right):=\frac{h\left(\check{\varphi}_{w}\left(s\right)\right)}{h\left(s\right)}. Consistency follows from harmonicity: \displaystyle\sum^{m}_{i=1}\mu_{s}\left(\left[wi\right]\right) \displaystyle=\sum^{m}_{i=1}\frac{h\left(\check{\varphi}_{wi}\left(s\right)\right)}{h\left(s\right)} \displaystyle=\frac{\sum^{m}_{i=1}h\left(\varphi_{i}\left(\check{\varphi}_{w}\left(s\right)\right)\right)}{h\left(s\right)}=\frac{h\left(\check{\varphi}_{w}\left(s\right)\right)}{h\left(s\right)}=\mu_{s}\left(\left[w\right]\right). Hence \mu_{s} defines a unique probability measure on \left(\Omega,\mathcal{F}\right) by Kolmogorov extension. The product formula ( 5.1 ) follows: \frac{h\left(\check{\varphi}_{w}\left(s\right)\right)}{h\left(s\right)}=\prod^{n}_{k=1}\frac{h\left(\check{\varphi}_{i_{1}\cdots i_{k}}\left(s\right)\right)}{h\left(\check{\varphi}_{i_{1}\cdots i_{k-1}}\left(s\right)\right)}=\prod^{n}_{k=1}p_{i_{k}}\left(\check{\varphi}_{i_{1}\cdots i_{k-1}}\left(s\right)\right). The Markov property is immediate from the cylinder specification. Indeed, conditioning on \mathcal{F}_{n} is conditioning on the prefix \omega|n , hence on the current state s_{n}=\check{\varphi}_{\omega|n}\left(s\right) , and the conditional distribution of \omega_{n+1} is given by p_{i}\left(s_{n}\right) , which is ( 5.2 ) . The identity ( 5.3 ) is the standard reformulation of this transition rule. ∎ It is convenient to isolate the Doob transformed averaging operator on functions induced by the measures \mu_{s} . Define, for bounded f on X^{+}_{\mathrm{fin}} , (5.4) \left(Qf\right)\left(s\right):=\sum^{m}_{i=1}p_{i}\left(s\right)f\left(\varphi_{i}\left(s\right)\right)=\frac{1}{h\left(s\right)}\sum^{m}_{i=1}h\left(\varphi_{i}\left(s\right)\right)f\left(\varphi_{i}\left(s\right)\right). Then Q is a Markov operator on functions on X^{+}_{\mathrm{fin}} , and it is intertwined with the original branching operator P by the gauge h . Lemma 5.2 . For every bounded f on X^{+}_{\mathrm{fin}} , P\left(hf\right)=h\left(Qf\right). Consequently, for every n\geq 0 , P^{n}\left(hf\right)=h\left(Q^{n}f\right),\qquad Q^{n}f\left(s\right)=\mathbb{E}_{\mu_{s}}\left[f\left(s_{n}\right)\right]. Proof. By definition (using ( 5.4 )), P\left(hf\right)\left(s\right)=\sum^{m}_{i=1}h\left(\varphi_{i}\left(s\right)\right)f\left(\varphi_{i}\left(s\right)\right)=h\left(s\right)\left(Qf\right)\left(s\right). Iterating gives P^{n}\left(hf\right)=h\left(Q^{n}f\right) . The formula Q^{n}f\left(s\right)=\mathbb{E}_{\mu_{s}}\left[f\left(s_{n}\right)\right] is the standard Markov chain identity from Proposition 5.1 . ∎ We now turn to kernels. The next object is the natural normalization of kernels by h , and the corresponding transform of L . Given a kernel J on X^{+}_{\mathrm{fin}}\times X^{+}_{\mathrm{fin}} we define its h -normalization by J^{\left(h\right)}\left(s,t\right):=\frac{J\left(s,t\right)}{h\left(s\right)h\left(t\right)}. We then define an operator \widetilde{L} on kernels by (5.5) (\widetilde{L}G)\left(s,t\right):=\sum^{m}_{i=1}p_{i}\left(s\right)p_{i}\left(t\right)G\left(\varphi_{i}\left(s\right),\varphi_{i}\left(t\right)\right). This operator is the kernel-level analogue of Q , except that the same symbol i is used in both variables, reflecting the fact that L branches synchronously in s and t . Proposition 5.3 . For every kernel J on X^{+}_{\mathrm{fin}}\times X^{+}_{\mathrm{fin}} , \left(LJ\right)^{\left(h\right)}=\widetilde{L}(J^{\left(h\right)}). In particular, for the kernel tower K_{n}=L^{n}K , K^{\left(h\right)}_{n}=\widetilde{L}^{n}(K^{\left(h\right)}),\qquad n\geq 0, and if K_{\infty} is L -invariant then K^{\left(h\right)}_{\infty} is \widetilde{L} -invariant. Proof. Compute directly: \displaystyle\left(LJ\right)^{\left(h\right)}\left(s,t\right) \displaystyle=\frac{\sum^{m}_{i=1}J\left(\varphi_{i}\left(s\right),\varphi_{i}\left(t\right)\right)}{h\left(s\right)h\left(t\right)} \displaystyle=\sum^{m}_{i=1}p_{i}\left(s\right)p_{i}\left(t\right)\frac{J\left(\varphi_{i}\left(s\right),\varphi_{i}\left(t\right)\right)}{h\left(\varphi_{i}\left(s\right)\right)h\left(\varphi_{i}\left(t\right)\right)}. The last expression is (\widetilde{L}(J^{\left(h\right)}))\left(s,t\right) ; see ( 5.5 ). Iteration gives K^{\left(h\right)}_{n}=\widetilde{L}^{n}\left(K^{\left(h\right)}\right) , and invariance passes through the same intertwining. ∎ The operator \widetilde{L} admits an explicit boundary expansion in terms of the Doob measures \mu_{s} . For a word w=i_{1}\cdots i_{n} we set p_{w}\left(s\right):=\mu_{s}\left(\left[w\right]\right)=\frac{h\left(\check{\varphi}_{w}\left(s\right)\right)}{h\left(s\right)}. Then p_{w}\left(s\right)\geq 0 and \sum_{|w|=n}p_{w}\left(s\right)=1 for each fixed n . Lemma 5.4 . For every kernel G on X^{+}_{\mathrm{fin}}\times X^{+}_{\mathrm{fin}} and every n\geq 0 , (\widetilde{L}^{n}G)\left(s,t\right)=\sum_{|w|=n}p_{w}\left(s\right)p_{w}\left(t\right)G\left(\check{\varphi}_{w}\left(s\right),\check{\varphi}_{w}\left(t\right)\right). Equivalently, if \mathcal{F}_{n} denotes the length- n cylinder \sigma -algebra on \Omega , then the finite measure on \Omega defined on cylinders by \nu^{\left(n\right)}_{s,t}\left(\left[w\right]\right):=p_{w}\left(s\right)p_{w}\left(t\right) satisfies (\widetilde{L}^{n}G)\left(s,t\right)=\int_{\Omega}G\left(\check{\varphi}_{\omega|n}\left(s\right),\check{\varphi}_{\omega|n}\left(t\right)\right)d\nu^{\left(n\right)}_{s,t}\left(\omega\right), where the integrand is \mathcal{F}_{n} -measurable. Proof. The identity is proved by induction on n . The case n=1 is the definition of \widetilde{L} , since p_{i}\left(s\right)=p_{\left(i\right)}\left(s\right) . Suppose the identity holds at level n and apply \widetilde{L} once more: (\widetilde{L}^{n+1}G)\left(s,t\right)=\sum^{m}_{i=1}p_{i}\left(s\right)p_{i}\left(t\right)(\widetilde{L}^{n}G)\left(\varphi_{i}\left(s\right),\varphi_{i}\left(t\right)\right). Insert the induction hypothesis at \varphi_{i}\left(s\right),\varphi_{i}\left(t\right) to obtain (\widetilde{L}^{n+1}G)\left(s,t\right)=\sum^{m}_{i=1}\sum_{|w|=n}p_{i}\left(s\right)p_{w}\left(\varphi_{i}\left(s\right)\right)p_{i}\left(t\right)p_{w}\left(\varphi_{i}\left(t\right)\right)G\left(\check{\varphi}_{wi}\left(s\right),\check{\varphi}_{wi}\left(t\right)\right). Using the multiplicative rule p_{wi}\left(s\right)=p_{w}\left(\varphi_{i}\left(s\right)\right)p_{i}\left(s\right) , we identify the coefficient as p_{wi}\left(s\right)p_{wi}\left(t\right) and recover the claimed sum over words of length n+1 . The integral form is just a rewriting of the same cylinder expansion. ∎ We next connect this boundary expansion to the defect decomposition from Section 2 . Recall the defect kernels D_{n}:=K_{n+1}-K_{n},\qquad n\geq 0, which are positive definite by monotonicity of the kernel tower. Writing D_{0}=LK-K , one has D_{n}=L^{n}D_{0} and hence, by Proposition 5.3 , D^{\left(h\right)}_{n}=\widetilde{L}^{n}(D^{\left(h\right)}_{0}). Therefore Lemma 5.4 yields an explicit boundary expansion for each normalized defect kernel: D^{\left(h\right)}_{n}\left(s,t\right)=\sum_{|w|=n}p_{w}\left(s\right)p_{w}\left(t\right)D^{\left(h\right)}_{0}\left(\check{\varphi}_{w}\left(s\right),\check{\varphi}_{w}\left(t\right)\right). Summing in n and using the telescoping identity K_{\infty}=K+\sum_{n\geq 0}D_{n} from Section 2 , we obtain a boundary organized representation for the normalized invariant completion: K^{\left(h\right)}_{\infty}\left(s,t\right)=K^{\left(h\right)}\left(s,t\right)+\sum^{\infty}_{n=0}\sum_{|w|=n}p_{w}\left(s\right)p_{w}\left(t\right)D^{\left(h\right)}_{0}\left(\check{\varphi}_{w}\left(s\right),\check{\varphi}_{w}\left(t\right)\right),\qquad s,t\in X^{+}_{\mathrm{fin}}. This formula is purely deterministic and already exhibits the boundary filtration by word length. We now include a Hilbert space realization of the same expansion as an L^{2} boundary Gram kernel; this is the boundary analogue of the direct-sum defect realization from Section 2 . Theorem 5.5 . Let H_{D^{\left(h\right)}_{0}} be the RKHS of the normalized defect kernel D^{\left(h\right)}_{0} on X^{+}_{\mathrm{fin}} . Define a Hilbert space H_{\partial}:=\ell^{2}\left(\mathbb{N}_{0}\right)\otimes L^{2}\left(\Omega,\nu\right)\otimes H_{D^{\left(h\right)}_{0}}, where \nu is any probability measure on \Omega for which cylinders have positive measure (for example the Bernoulli measure). For each s\in X^{+}_{\mathrm{fin}} define \Psi\left(s\right)\in H_{\partial} by the rule that its (n,\omega) -fiber equals \Psi\left(s\right)\left(n,\omega\right)=\frac{p_{\omega|n}\left(s\right)}{\sqrt{\nu\left(\left[\omega|n\right]\right)}}(D^{\left(h\right)}_{0})_{\check{\varphi}_{\omega|n}\left(s\right)}\in H_{D^{\left(h\right)}_{0}},\qquad n\geq 0, where (D^{\left(h\right)}_{0})_{x} denotes the kernel section at x in H_{D^{\left(h\right)}_{0}} . Then for all s,t\in X^{+}_{\mathrm{fin}} , \left\langle\Psi\left(s\right),\Psi\left(t\right)\right\rangle_{H_{\partial}}=\sum^{\infty}_{n=0}D^{\left(h\right)}_{n}\left(s,t\right). Consequently, K^{\left(h\right)}_{\infty}\left(s,t\right)=K^{\left(h\right)}\left(s,t\right)+\left\langle\Psi\left(s\right),\Psi\left(t\right)\right\rangle_{H_{\partial}}. Proof. Fix s,t\in X^{+}_{\mathrm{fin}} . By construction and orthogonality in \ell^{2}\left(\mathbb{N}_{0}\right) , \left\langle\Psi\left(s\right),\Psi\left(t\right)\right\rangle_{H_{\partial}}=\sum^{\infty}_{n=0}\int_{\Omega}\left\langle\Psi\left(s\right)\left(n,\omega\right),\Psi\left(t\right)\left(n,\omega\right)\right\rangle_{H_{D^{\left(h\right)}_{0}}}\,d\nu\left(\omega\right). For fixed n , the integrand is constant on cylinders \left[w\right] with |w|=n , and on such a cylinder we have \omega|n=w . Using \left\langle(D^{\left(h\right)}_{0})_{x},(D^{\left(h\right)}_{0})_{y}\right\rangle=D^{\left(h\right)}_{0}\left(x,y\right) , we obtain \displaystyle\int_{\Omega}\left\langle\Psi\left(s\right)\left(n,\omega\right),\Psi\left(t\right)\left(n,\omega\right)\right\rangle d\nu\left(\omega\right) \displaystyle=\sum_{|w|=n}\frac{p_{w}\left(s\right)p_{w}\left(t\right)}{\nu\left(\left[w\right]\right)}D^{\left(h\right)}_{0}\left(\check{\varphi}_{w}\left(s\right),\check{\varphi}_{w}\left(t\right)\right)\nu\left(\left[w\right]\right) \displaystyle=\sum_{|w|=n}p_{w}\left(s\right)p_{w}\left(t\right)D^{\left(h\right)}_{0}\left(\check{\varphi}_{w}\left(s\right),\check{\varphi}_{w}\left(t\right)\right). By Lemma 5.4 applied to G=D^{\left(h\right)}_{0} , the right-hand side equals \left(\widetilde{L}^{n}D^{\left(h\right)}_{0}\right)\left(s,t\right)=D^{\left(h\right)}_{n}\left(s,t\right) . Summing over n gives \left\langle\Psi\left(s\right),\Psi\left(t\right)\right\rangle_{H_{\partial}}=\sum^{\infty}_{n=0}D^{\left(h\right)}_{n}\left(s,t\right). Finally, K^{\left(h\right)}_{\infty}=K^{\left(h\right)}+\sum_{n\geq 0}D^{\left(h\right)}_{n} is the normalized form of the telescoping identity from Section 2 , and this yields the claimed representation. ∎ Remark 5.6 . The map s\mapsto\Psi\left(s\right) produces a boundary indexed family of features built from the one-step defect D_{0} propagated along the tree, with weights dictated by the Doob path measures. The \ell^{2}\left(\mathbb{N}_{0}\right) factor records the boundary filtration by word length, and the orthogonality across levels matches the defect splitting from Section 2 . In particular, the boundary representation above is canonical up to the choice of a reference cylinder-positive measure \nu on \Omega , and it is compatible with the Gaussian model of Section 4 in the sense that it gives an explicit boundary feature space whose Gram kernel is the accumulated defect \sum_{n\geq 0}D^{\left(h\right)}_{n} . We do not consider boundary limits along \omega\mapsto s_{n}\left(\omega\right) here, but the path measures \mu_{s} and the boundary feature map \Psi provide a convenient platform for such refinements, including martingale and boundary compactification questions, as well as for constructing additional L -invariant kernels by altering the defect contribution level-by-level along \Omega .