On domination for (non-symmetric) Dirichlet forms

Consider an open set $\Omega\subset{\mathbb{R}}^{n}$. Let $\Delta^{D}$ and $\Delta^{N}$ denote the Laplace operators on $L^{2}(\Omega)$ subject to the Dirichlet boundary condition $u|_{\partial\Omega}=0$ and the Neumann boundary condition $\frac{\partial u}{\partial n}|_{\partial\Omega}=0$, respectively. Here, $L^{2}(\Omega)$ denotes the space of all square-integrable functions on $\Omega$, $\partial\Omega$ represents the boundary of $\Omega$, and $\frac{\partial u}{\partial n}$ denotes the exterior normal derivative on the boundary (whose existence requires certain regularity assumptions). Denote by $(e^{t\Delta^{D}})_{t\geq 0}$ and $(e^{t\Delta^{N}})_{t\geq 0}$ the strongly continuous semigroups generated by $\Delta^{D}$ and $\Delta^{N}$, respectively. In [4], Arendt and Warma investigated the question of which self-adjoint operator generates a semigroup $(T_{t})_{t\geq 0}$ that is “sandwiched” between $(e^{t\Delta^{D}})_{t\geq 0}$ and $(e^{t\Delta^{N}})_{t\geq 0}$ in the sense that

where $\mathrm{p}L^{2}(\Omega)$ represents the family of all positive functions in $L^{2}(\Omega)$. Their main result demonstrates that, under certain necessary conditions, $(T_{t})_{t\geq 0}$ corresponds to the Laplace operator subject to a Robin boundary condition determined by a certain admissible measure $\mu$ on $\partial\Omega$ (see Example 7.6). This operator and its semigroup are denoted by $\Delta_{\mu}$ and $e^{t\Delta_{\mu}}$, respectively. Specifically, when $\mu$ is absolutely continuous with respect to the surface measure on $\partial\Omega$ with density $\beta$, $\Delta_{\mu}$ corresponds to the classical Robin boundary condition:

Arendt and Warma’s research builds upon Ouhabaz’s work [24], which provides a quadratic form characterization of the domination relation in (1.1). Specifically, if $(a^{1},\mathcal{D}(a^{1}))$ and $(a^{2},\mathcal{D}(a^{2}))$ are the closed forms associated with the strongly continuous semigroups $(T^{1}_{t})_{t\geq 0}$ and $(T^{2}_{t})_{t\geq 0}$, and both semigroups are positive (i.e., for any non-negative $f\in L^{2}(\Omega)$, $T^{1}_{t}f\geq 0$ and $T^{2}_{t}f\geq 0$), then

is equivalent to the conditions that $\mathcal{D}(a^{1})$ is an ideal of $\mathcal{D}(a^{2})$ and that

where $\mathrm{p}\mathcal{D}(a^{1})$ denotes the set of non-negative elements in $\mathcal{D}(a^{1})$ (see Lemma 2.1 for details).

In the context of (1.1), since the semigroups on both sides are positive and satisfy the (sub-)Markovian property (i.e., for any $f\in L^{2}(\Omega)$ with $0\leq f\leq 1$, it holds that $0\leq e^{t\Delta^{D}}f,e^{t\Delta^{N}}f\leq 1$), it follows that the sandwiched semigroup is also positive and Markovian. According to the foundational work of Beurling and Deny [6], semigroups with these properties correspond to closed forms known as Dirichlet forms.

Over the past two decades, the study of sandwiched semigroups and quadratic forms has been advanced and applied from various perspectives. To illustrate, we highlight a selection of contributions. While much of the research continues to focus on Laplace operators (or elliptic differential operators), the scope of boundary conditions has expanded to include non-local settings, as explored in [2, 25, 35]. Additionally, the focus has broadened to encompass general Dirichlet forms, with notable works including [5, 18, 22, 26, 29]. Recent years have also seen growing interest in extending these results to nonlinear semigroups and nonlinear Dirichlet forms, as exemplified by [11]. Beyond these theoretical advances, related research has found applications in areas such as partial differential equations, fractal structures, and other fields, as demonstrated in [8, 17, 28].

It is important to note that, under conditions of regularity or quasi-regularity, the relationship between Dirichlet forms and Markov processes in probability theory is well established (see [9, 15, 20]). However, to our knowledge, aside from the mention of the probabilistic counterparts of Dirichlet forms in [1], the aforementioned studies utilizing Ouhabaz’s domination criterion have primarily adopted an analytical perspective, largely overlooking the substantial potential of probabilistic methods, despite most discussions occurring within the framework of regularity. Researchers familiar with Markov processes can readily relate the domination relationship of $L^{2}$-semigroups from (1.2) to the subordination between two Markov transition semigroups: $(Q_{t})_{t\geq 0}$ is said to be subordinated to $(P_{t})_{t\geq 0}$ if

This represents a classic topic in the theory of general Markov processes (see [7, III] and [30, VII]). In simpler terms, the subordination stated in (1.3) corresponds to the killing transformation of Markov processes. The Markov process corresponding to $(Q_{t})_{t\geq 0}$, often referred to as the subprocess, is always derived from the Markov process corresponding to $(P_{t})_{t\geq 0}$ through a mechanism governed by some multiplicative functional that terminates sample paths. The Dirichlet form characterization of general killing transformations has also been addressed in prior studies (such as [38, 37]). Nevertheless, it is crucial to acknowledge that there exists a significant disparity between domination in the sense of almost everywhere from (1.1) and subordination in the pointwise sense from (1.3). Fortunately, the theory of regular Dirichlet forms established by Fukushima, along with the quasi-regular Dirichlet form framework developed by Ma et al. (see [20]), is specifically designed to bridge this gap. Therefore, at least theoretically, there exists a probabilistic pathway to investigate domination in (1.2).

The primary objective of this paper is precisely to establish the equivalence between the domination (1.2) of $L^{2}$-semigroups in analysis and the killing transformation of Markov processes in probability. Our principal result demonstrates the following intuitive fact: each dominated Dirichlet form is derived from two successive killing transformations. The first step involves terminating the Markov process when it exits a specific finely open set, which defines the actual state space of the subprocess corresponding to the dominated Dirichlet form. The second step consists of an additional killing transformation dictated by another multiplicative functional within the finely open set, which does not further alter the state space of the subprocess. At the same time, these two steps of killing transformations correspond to the analytical decomposition of domination described in Theorem 2.4: Every domination can be expressed as a composition of a Silverstein extension and strong subordination.

Extending beyond Ouhabaz’s domination criterion, our probabilistic characterization offers a representation of the dominated Dirichlet form using the bivariate Revuz measure of the multiplicative functional (which can be naturally extended to represent the sandwiched Dirichlet form, see Theorem 6.2). Specifically, for any (non-symmetric) Dirichlet forms $(a^{1},\mathcal{D}(a^{1}))$ and $(a^{2},\mathcal{D}(a^{2}))$ satisfying (1.2), there exists a positive measure $\sigma$ on $\overline{\Omega}\times\overline{\Omega}$, which can be expressed as detailed in (3.8), such that

where $\bar{\sigma}$ denotes the marginal measure of $\sigma$. It is noteworthy that [5] has achieved a similar representation for regular and symmetric Dirichlet forms using analytical methods.

Our investigation is situated within the general framework of non-symmetric Dirichlet forms, with the essential condition being the quasi-regularity of the dominated Dirichlet form (see §5). In other words, in the representation provided in (1.4), the quasi-regularity of $(a_{1},\mathcal{D}(a_{1}))$ is necessary, whereas $(a_{2},\mathcal{D}(a_{2}))$ does not need to satisfy the quasi-regularity condition. Consequently, the findings presented in this paper are applicable to the majority of situations discussed in the existing literature, including works such as [1, 5, 18, 22, 26, 29], among others. In particular, we re-examine the Laplacian-related sandwiched Dirichlet forms in §7. Theorem 7.3 eliminates all assumptions imposed by Arendt and Warma [4] concerning the characterization of sandwiched semigroups in (1.1): In fact, every semigroup $(T_{t})_{t\geq 0}$ satisfying (1.1) is determined by some quasi-admissible measure $\mu$ associated with the Robin Laplacian. (The definition of quasi-admissibility can be found in Definition 7.1.) Moreover, comprehensive solutions can also be provided for Laplacian-related problems in the non-local Robin boundary case. Notably, our results address an open problem posed in the literature [25], as discussed in §7.3.

Finally, we provide a brief description of commonly used symbols. The concepts and notations associated with Dirichlet forms align with those presented in [9, 15, 20], among others. For a function $u$ in a Dirichlet space, $\tilde{u}$ denotes its quasi-continuous modification by default. Given a topological space $E$, $\mathscr{B}(E)$ represents the collection of Borel measurable sets or Borel measurable functions, and $\mathscr{B}^{*}(E)$ denotes the collection of all universally measurable sets or universally measurable functions. For a class of functions $\mathscr{C}$, $\mathrm{p}\mathscr{C}$ and $\mathrm{b}\mathscr{C}$ denote the families of non-negative elements and bounded elements, respectively, within $\mathscr{C}$.

Let $E$ be a measurable space, and let $m$ be a $\sigma$-finite measure on $E$. The norms of $L^{2}(E,m)$ and $L^{\infty}(E,m)$ are denoted by $\|\cdot\|_{2}$ and $\|\cdot\|_{\infty}$, respectively, while the inner product of $L^{2}(E,m)$ is denoted by $(\cdot,\cdot)_{m}$. A (non-symmetric) Dirichlet form $({\mathscr{E}},{\mathscr{F}})$ on $L^{2}(E,m)$ is defined as a coercive closed form that satisfies the Markovian property, as described in, e.g., [20, I, Definition 4.5]. The ${\mathscr{E}}_{1}$-norm on ${\mathscr{F}}$ is denoted by $\|\cdot\|_{{\mathscr{E}}_{1}}$, where $\|f\|_{{\mathscr{E}}_{1}}:=\left({\mathscr{E}}(f,f)+(f,f)_{m}\right)^{1/2}$ for all $f\in{\mathscr{F}}$. The Dirichlet form is called symmetric if, in addition, ${\mathscr{E}}(f,g)={\mathscr{E}}(g,f)$ for all $f,g\in{\mathscr{F}}$.

For a given Dirichlet form $({\mathscr{E}},{\mathscr{F}})$ on $L^{2}(E,m)$, it is well known that $\mathrm{b}{\mathscr{F}}:={\mathscr{F}}\cap L^{\infty}(E,m)$ is a subalgebra of $L^{\infty}(E,m)$ (see [20, I, Corollary 4.15]), and ${\mathscr{F}}$ forms a sublattice of $L^{2}(E,m)$ (see [20, I, Proposition 4.11]).

If $U,V$ are subalgebras of $L^{\infty}(E,m)$, we say that $U$ is an algebraic ideal in $V$ if $f\in U$ and $g\in V$ imply $fg\in U$. For sublattices $U,V$ of $L^{2}(E,m)$, $U$ is called an order ideal in $V$ if $f\in U,g\in V$, and $|g|\leq|f|$ imply $g\in U$.

For two Dirichlet forms $({\mathscr{E}},{\mathscr{F}})$ and $({\mathscr{E}}^{\prime},{\mathscr{F}}^{\prime})$, we say that ${\mathscr{E}}$ dominates ${\mathscr{E}}^{\prime}$ (or equivalently, ${\mathscr{E}}^{\prime}$ is dominated by ${\mathscr{E}}$) and write $({\mathscr{E}}^{\prime},{\mathscr{F}}^{\prime})\preceq({\mathscr{E}},{\mathscr{F}})$ or ${\mathscr{E}}^{\prime}\preceq{\mathscr{E}}$ if either of the conditions in the following lemma is satisfied. The second condition in this lemma is commonly referred to as the Ouhabaz domination criterion (see [24]) for the domination of semigroups.

We introduce two additional notions.

Clearly, ${\mathscr{E}}^{\prime}\preceq{\mathscr{E}}$ if and only if ${\mathscr{E}}^{\prime}$ is subordinate to ${\mathscr{E}}$ and $\mathrm{b}{\mathscr{F}}^{\prime}$ is an algebraic ideal in $\mathrm{b}{\mathscr{F}}$ (or equivalently, ${\mathscr{F}}^{\prime}$ is an order ideal in ${\mathscr{F}}$). By the proof of Lemma 2.1, ${\mathscr{E}}$ is a Silverstein extension of ${\mathscr{E}}^{\prime}$ if and only if ${\mathscr{F}}^{\prime}$ is an order ideal in ${\mathscr{F}}$ and ${\mathscr{E}}={\mathscr{E}}^{\prime}$ on $\mathrm{b}{\mathscr{F}}^{\prime}\times\mathrm{b}{\mathscr{F}}^{\prime}$. In particular, if ${\mathscr{E}}$ is a Silverstein extension of ${\mathscr{E}}^{\prime}$, then ${\mathscr{E}}^{\prime}\preceq{\mathscr{E}}$ and ${\mathscr{E}}^{\prime}$ is subordinate to ${\mathscr{E}}$.

We now present a simple yet intriguing result, which demonstrates that domination can be uniquely decomposed into a combination of strong subordination and Silverstein extension.

The relationship between Dirichlet forms and Markov processes is commonly known as contingent upon the condition of quasi-regularity. In Appendix A, we present an overview of the content relevant to this condition and the properties of the corresponding Markov processes.

Let $E$ be a Lusin topological space and $m$ a $\sigma$-finite measure on $E$ with support $\text{supp}[m]=E$. We consider a quasi-regular Dirichlet form $({\mathscr{E}},{\mathscr{F}})$ on $L^{2}(E,m)$. Let $X=\left(\Omega,\mathcal{F},\mathcal{F}_{t},X_{t},\theta_{t},\mathbb{P}_{x}\right)$ with lifetime $\zeta$ be an $m$-tight Borel right process properly associated with $({\mathscr{E}},{\mathscr{F}})$. The cemetery is denoted by $\Delta$. We may assume that $\Omega$ contains a distinguished point $[\Delta]$ such that $X_{t}([\Delta])=\Delta$ for all $t\geq 0$. The goal of this section is to demonstrate that the killing transformation on $X$ induces the domination of $({\mathscr{E}},{\mathscr{F}})$.