Abstract
We consider the limit of solutions of scaled linear kinetic equations with a reflection-transmission-killing condition at the interface. Both the coefficient describing the probability of killing and the scattering kernel degenerate. We prove that the long-time, large-space limit is the unique solution of a version of the fractional in space heat equation that corresponds to the Kolmogorov equation for a symmetric stable process, which is reflected, or transmitted while crossing the interface and is killed upon the first hitting of the interface. The results of the paper are related to the work in Komorowski et al. (Ann Prob 48:2290–2322, 2020), where the case of a non-degenerate probability of killing has been considered.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
We study the asymptotic behaviour of a linear kinetic equation on the real line \({\mathbb {R}}\):
with the following boundary condition at \(x=0\) and \(t\ge 0\):
Here, \({\mathbb {T}}\) denotes the unit one-dimensional torus, understood as the interval \([-1/2,1/2]\) with identified endpoints, \({\mathbb {T}}_\pm :=[k\in {\mathbb {T}}:\pm k>0]\) and \({\mathbb {R}}_*:={\mathbb {R}}\setminus \{0\}\). We call the origin \(o:=[y=0]\) interface. The parameters \(\gamma >0\), \(T_o\ge 0\) are given and the functions \(\bar{\omega }, p_\pm , p_0\), defined on \({\mathbb {T}}\), are assumed to be continuous and non-negative. The scattering operator \(L_k\), acting only on the variable k in \({\mathbb {T}}\), is given by
Here \(R:{\mathbb {T}}^2\rightarrow [0,+\infty )\) is \(C^2\) smooth and u belongs to \(L^1({\mathbb {T}})\).
This equation arises in the kinetic limit of the evolution of the energy density in a stochastically perturbed harmonic chain interacting with a point Langevin thermostat located at \(y=0\), see [22,23,24,25]. The energy density W(t, y, k) at time t is resolved in both the spatial variable \(y\in {\mathbb {R}}\) and the frequency variable \(k\in {\mathbb {T}}\). The function \(\bar{\omega }\) is the dispersion relation of the harmonic chain and \(\bar{\omega }'(k)\) is the group velocity of phonons of mode k – theoretical particles that carry the energy due to the chain vibrations of frequency k. The presence of the Langevin thermostat results in the boundary (interface) condition (1.2). The number \(T_o\) corresponds to the temperature of the thermostat, while \(p_+(k)\), \(p_-(k)\) and \(p_0(k)\) are the respective probabilities of transmission, reflection and killing of a mode k phonon at the interface; see [23]. They are continuous, even functions that satisfy
With a small risk of ambiguity, in what follows we also write
Our main goal is to study the asymptotic behaviour of the solutions of (1.1) on macroscopic space-time scales. More precisely, we are interested in the limit of solutions, when \(\lambda \rightarrow +\infty \), for the family of rescaled equations
subject to the boundary condition (1.2), where \(1<\alpha <2\) is a suitably chosen exponent.
It turns out, see [5, 22], that under appropriate assumptions on the total scattering kernel \(R(k):=\int _{{\mathbb {T}}}R(k,k')dk'\), group velocity \(\bar{\omega }'(k)\) and a suitable choice of \(\alpha \), the limit \(\bar{W}(t,y):=\lim _{\lambda \rightarrow +\infty }W_\lambda (t,y,k)\) exists in a weak (distributional) sense. In the case where \(\sup _{k\in {\mathbb {T}}}[\bar{\omega }'(k)]^2/R(k)<+\infty \) and \(\alpha =2\) (diffusive space-time scaling), \(\bar{W}(t,y)\) satisfies the heat equation with the Dirichlet boundary condition \(\bar{W}(t,0)=T_o\) and \(\bar{W}(0,y)=\int _{{\mathbb {T}}}W(0,y,k)dk\), see [5, Theorem 2.2]. On the other hand, if \(|\bar{\omega }'(k)|/R(k)\sim |k|^{-\beta }\), for \(|k|\ll 1\), with \(\beta >1\) and \(p_0=\lim _{k\rightarrow 0}p_0(k)>0\) (this is, e.g., the case of an unpinned nearest neighbor harmonic chain, see [25, formula (119)]), then letting \(\alpha =1+1/\beta \) one can show, see [22, Theorem 1.1], that \(\bar{W}(t,y)\) satisfies a fractional diffusion equation with a boundary condition at \(y=0\). In informal terms, the equation states that
where \({\mathcal L}\) is the generator of a modified symmetric \((1+1/\beta )\)-stable process. The process behaves like a “regular” symmetric \((1+1/\beta )\)-stable process outside the interface \(y=0\), but transmits, reflects, or dies at the interface with the respective probabilities \(p_+\), \(p_-\) and \(p_0\) of (1.5).
In the present paper, we address the situations where \(p_0=0\), in fact, we assume that the killing probability \(p_0(k)\) satisfies, for some \(\kappa >0\),
Furthermore, we suppose that the transmission probability does not vanish, that is,
The dispersion relation \(\bar{\omega }:{\mathbb {T}}\rightarrow [0,+\infty )\) is assumed to be even and unimodal, i.e., it possesses exactly one local maximum, at 1/2, and one local minimum, at 0. In addition it is of the \(C^2({\mathbb {T}}_*)\) class of regularity, where \({\mathbb {T}}_*:={\mathbb {T}}\setminus \{0\}\), and there exist one-sided limits of its both derivatives at \(k=0\). A typical dispersion relation we have in mind is \(\bar{\omega }(k)=|\sin (\pi k)|\), which corresponds to the harmonic chain with the nearest-neighbour interaction.
Concerning the scattering kernel, we assume that it is of the multiplicative form
where \(R_1\), \(R_2\) are non-negative, even functions belonging to \(C({\mathbb {T}})\). Without loss of generality, we also suppose that
We suppose that there exist exponents \(\beta _j>0\), \(j=1,2,3\), such that
Consistent with the condition (1.13), we further require that
We define the probability measure \(\pi (dk):=R_2(k)/\big ({\mathcal R} R_1(k)\big )dk\) and we can informally state the main result of the present work; see Theorem 2.5 for a rigorous formulation.
Main Theorem. Let \(W_0(y,k)\) be a sufficiently regular function satisfying the interface conditions in (1.2) and such that the averaged function
satisfies \(\bar{W}_0(0)=T_o\). Let \(W_\lambda (t,y,k)\) be the solution to (1.6) with interface conditions (1.2). Then, under the hypotheses (1.8)–(1.14) made above, the limit \( \lim _{\lambda \rightarrow +\infty }W_\lambda (t,y,k) \, = \, \bar{W}(t,y)\) exists in a distributional sense on \({\mathbb {R}}\times {\mathbb {T}}\) for any \(t>0\). Moreover, \(\bar{W}(t,y)\) is the unique weak solution, in the sense of Definition 2.4 below, of the equation:
with the initial and boundary conditions \(\bar{W}(0,y) = \bar{W}_0(y)\) and \(\bar{W}(t,0)\, = \, T_o\), respectively. Here, \(\bar{\gamma }\) is a positive constant depending on the parameters of the model (see (2.9) below),
and \(\Gamma (\cdot )\) is the Euler gamma function.
Remark 1.1
The above result should be compared with the result of [22], where the case \(p_0=\lim _{k\rightarrow 0}p_0(k)>0\) has been considered. As we have already mentioned, the informal formulation of the limiting fractional dynamics involves the generator \({\mathcal L}\) of the process that is symmetric stable and can be transmitted, reflected or killed when crossing the interface at \(y=0\). In the situation considered in the present paper we can also view the evolution of \(\bar{W}(t,y)\) as described by Eq. (1.7), with \({\mathcal L}\) corresponding to a symmetric \(\alpha \)-stable process that is transmitted or reflected when crossing the interface at \(y=0\), with the respective probabilities \(p_+\) and \(p_-\). Furthermore, the process is killed upon the first hitting of the interface.
The fractional diffusion limit of a kinetic equation has been the subject of intense investigation in recent years. We refer the interested reader to a review of the existing literature contained in [26]. However, there seem to be only few results dealing with a fractional diffusion limit for kinetic equations with a boundary condition. In this context, we mention the papers [1, 3, 9,10,11,12, 22]. The case that is somewhat related to ours is considered in [11] and [12]. In the first paper, the convergence of scaled solutions to kinetic equations in spatial dimension one, with diffusive reflection condition on the boundary, is investigated. However, this condition is different from ours. Furthermore, the results of [11] do not establish the uniqueness of the limit, stating only that it satisfies a certain fractional diffusive equation with a boundary condition and leaving the question of the uniqueness of solutions for the limiting equation open, see the remark after Theorem 1.2 in [11]. In the paper [12], the authors complete the results of [11] and establish an anomalous diffusive limit for a family of solutions of scaled linear kinetic equations in a one-dimensional bounded domain with diffusive boundary conditions. The scattering kernel, defined on \({\mathbb {R}}^2\), is also of multiplicative form, as in (1.10), and decays according to an appropriate power law.
Concerning our methods of proof, we rely on the probabilistic interpretation of solutions to kinetic equations. As shown in [22, Proposition 3.2], the solution can be expressed, with the help of an underlying two-dimensional stochastic process \(\{\big (Y^o(t),K^o(t)\big )\}_{t\ge 0}\), see Proposition 3.1 below. The component \(K^o(t)\) shall be referred to as a frequency, while \(Y^o(t)\) can be thought of as the position of a particle, whose velocity is given by \(-\bar{\omega }'(K^o(t))\), see (3.7) below. Outside the interface \([y=0]\), i.e. when \(Y^o(t)\not =0\), the component \(K^o(t)\) is described by a pure jump, \({\mathbb {T}}\)-valued Markov process corresponding to the operator \(L_k\) in (1.3). As a result, the particle performs a uniform motion between the consecutive scattering events, with velocity \(-\bar{\omega }'(K^o(t))\). On the other hand, if the particle tries to cross the interface at time t, it will be transmitted (then \(K^o(t)=K^o(t-)\)), reflected (then \(K^o(t)=-K^o(t-)\)), or killed, with the respective probabilities \(p_+(K^o(t-))\), \(p_-(K^o(t-))\) and \(p_0(K^o(t-))\). Using this probabilistic interpretation, the asymptotic behaviour of the solutions of (1.6) can be reformulated into the problem of finding the limit of appropriately scaled processes \(\{\big (Y_\lambda ^o(t),K_\lambda ^o(t)\big )\}_{t\ge 0}\), with the parameter \(\lambda >0\) corresponding to the ratio between the macro- and microscopic time units. This approach has been successfully implemented in the case of solutions of scaled kinetic equations without interface in [4, 19, 20]. In the case when the killing probability \(p_0\) is strictly positive, as assumed in [22], one can expect that typically the particle crosses the interface only finitely many times before being killed. Its trajectory can then be obtained by a path transformation (consisting in performing suitable reflections) of the trajectory when no interface is present. This transformation turns out to be continuous in the Skorokhod \(J_1\) topology on the path space, thus the convergence of the scaled processes in the presence of the interface is a consequence of the result for the process in the free space, done, e.g., in [20]. This approach cannot be applied in our present situation, since with small \(p_0\) it is not so obvious how to effectively control the number of interface crossings performed by the particle before it gets killed. For this reason, we use a different method to deal with the problem. We define a Lévy-type process \(\big \{Z(t)\big \}_{t\ge 0}\) whose scaled limit \(\big \{Z_{\lambda }(t)\big \}_{t\ge 0}\), as \(\lambda \rightarrow +\infty \), is related to the respective limit of the position of the particle by a deterministic time change. To find the limit of \(\big \{Z_{\lambda }(t)\big \}_{t\ge 0}\), we prove that the associated Dirichlet forms converge in the sense of Mosco (see [27, Definition 2.1.1] and also Definition 6.1 below), to the Dirichlet form corresponding to the modified \(\alpha \)-stable process, described in Remark 1.1. The method used in Sect. 6 can also be applied to the case considered in [22]. However, one should keep in mind that [22] covers a broader range of scattering kernels than the multiplicative ones considered here, see (1.10). In fact, it is this particular form of the scattering kernel that allows us to use the machinery of Dirichlet forms in the present paper.
Apart from the limit theory of kinetic equations, we present here some new technical results, which may be of independent interest, for a construction of skew-stable processes (Theorem 3.4), for the characterization of the fractional Sobolev spaces (Proposition 6.3 and Corollary 6.4) and for the definition and uniqueness of solutions to non-local parabolic equations (Definition 2.4 and Proposition 3.7).
The paper is organized as follows: after presenting some preliminaries in Sect. 2, we reformulate the problem of finding the limit of solutions of (1.6), with the boundary condition (1.2) into the problem of finding the limit of the respective stochastic processes. This is done in Sect. 3. In Sect. 4, we show how to use the result concerning the limit of stochastic processes to conclude our main result, formulated rigorously in Theorem 2.5. Section 5 is devoted to the proof of the convergence of the scaled Lévy type processes, as stated in Theorem 3.4, assuming the convergence of their respective Dirichlet forms in the sense of Mosco. The latter is established in Sect. 6. Some auxiliary facts are proved in “Appendix A and B”.
2 Preliminaries
2.1 Precise assumptions on the model
Given an arbitrary set A and two functions \(f,g:A\rightarrow [0,+\infty ]\), we write \(f\preceq g\) on A if there is a constant \(C>0\) such that
We also write \(f\approx g\) on A when \(f\preceq g\) and \(g\preceq f\). In what follows, we always assume the following conventions: \(\sum _A f(a) =0\) and \(\prod _A f(a)=1\), if \(A=\emptyset \), and \(\bar{\omega }'(0):=0\), even though as a rule, \(\bar{\omega }\) is not differentiable at \(k=0\).
Let us now denote for any k in \({\mathbb {T}}_*\),
which can be interpreted as expected waiting time for the scattering of a particle at frequency k. The interplay between scattering kernel and drift will be captured by the function
Clearly, we can interpret |S(k)| as the expected distance travelled by the particle before scattering. According to our assumptions, S is an odd function and by (1.12), \(\lim _{k\rightarrow 0 } |k|^{\beta _3} |S(k)|=S_*\). We impose some further restrictions on the model coefficients by assuming that:
Note that assumption made about \(S'\) in third line of (2.3) together with the fact that \(\bar{\omega }'(k)\ge 0\), for \(k\in (0,1/2]\) imply that S restricted to that interval is strictly decreasing. From this point on, we shall assume that all the above hypotheses together with those made including and between (1.8) and (1.15) are in force.Footnote 1
Remark 2.1
It follows from our subsequent argument that the limiting behavior of \(W_\lambda (t,y,k)\) is determined by the behavior of S(k) and \(R_j(k)\), \(j=1,2\), in the vicinity of \(k=0\), provided that these functions, together with \(1/R_1(k)\) remain bounded for k away from 0. The generalization of the proof for a broader class of coefficients should therefore be possible, at the expense of an additional complication of the argument.
2.2 Solution of the kinetic equation
Let us write \({\mathbb {R}}_+:=(0, +\infty )\), \(\bar{{\mathbb {R}}}_+:=[0,+\infty )\), \({\mathbb {R}}_-:=(-\infty ,0)\) and \(\bar{{\mathbb {R}}}_-:=(-\infty , 0]\). Given \(T \in {\mathbb {R}}\), we denote by \(\mathcal {C}_{T}\) the class of functions \(\phi \) in \(C_b({\mathbb {R}}_*\times {\mathbb {T}}_*)\) that can be continuously extended to \(\bar{{\mathbb {R}}}_\iota \times {\mathbb {T}}_*\), for \(\iota \in \{+,-\}\), and satisfy the following interface conditions:
Clearly, the constant function \(T_o\) belongs to \(\mathcal {C}_{T_o}\). Furthermore, \(F\in \mathcal { C}_{T_o}\) if and only if \(F-T_o\in \mathcal {C}_0\). This allows us to reduce the proofs of some results below to the case \(T_o=0\).
Remark 2.2
Note that if \(\phi \in \mathcal { C}_{T_o}\) and \( \phi (y,k)=\tilde{\phi }(y)\) for all \((y,k)\in {\mathbb {R}}_*\times {\mathbb {T}}_*\), then, we necessarily have \(\tilde{\phi }(0^+)=\tilde{\phi }(0^-)=T_o\). In light of the previous remark, it suffices to verify the statement for \(T_o=0\). Taking into account (2.4) (with \(T_o=0\)) we conclude that
The form \((x,y)\mapsto (1-p_-(k))\left( x^2+y^2\right) -2 p_-(k)p_+(k) xy\) is positive definite, since
due to (1.9).
Following [22], we now recall the definition of solution to equation (1.1) with the interface conditions (1.2).
Definition 2.3
A bounded, continuous function \(W:{\mathbb {R}}_+\times {\mathbb {R}}_*\times {\mathbb {T}}_*\rightarrow {\mathbb {R}}\) is called a solution to equation (1.1) with the interface conditions (1.2) if all the following conditions are satisfied:
-
for any \(\iota , \iota '\) in \(\{-,+\}\), the restriction of W to \({\mathbb {R}}_+\times {\mathbb {R}}_\iota \times {\mathbb {T}}_{\iota '}\) can be extended to a bounded, continuous function on \(\bar{\mathbb {R}}_+\times \bar{\mathbb {R}}_{\iota }\times \bar{\mathbb {T}}_{\iota '}\);
-
for any \((t,y,k)\in {\mathbb {R}}_+\times {\mathbb {R}}_*\times {\mathbb {T}}_*\) fixed, the function \(s\mapsto W(t+s,y+\bar{\omega }'(k) s,k)\) is continuously differentiable in a neighborhood of \(s=0\), the directional derivative
$$\begin{aligned} D_tW(t,y,k)\,:= \, \frac{d}{ds}_{|s=0} W(t+s,y+\bar{\omega }'(k) s,k) \end{aligned}$$(2.5)is bounded in \({\mathbb {R}}_+\times {\mathbb {R}}_*\times {\mathbb {T}}_*\), and
$$\begin{aligned} D_tW(t,y,k) \, = \, \gamma L_k W(t,y,k), \quad (t,y,k)\in {\mathbb {R}}_+\times {\mathbb {R}}_*\times {\mathbb {T}}_*; \end{aligned}$$ -
the interface conditions in (1.2) hold, together with the initial condition
$$\begin{aligned} \lim _{t\rightarrow 0+} W(t,y,k)\, = \, W_0(y,k),\quad (y,k)\in {\mathbb {R}}_*\times {\mathbb {T}}_*. \end{aligned}$$
We highlight that, at least formally, the directional derivative satisfies
which justifies the notation in (2.5). It has been shown in [22] that there exists a unique (classical) solution, in the sense of Definition 2.3, to the Cauchy problem (1.1), if the initial distribution \(W_0\) belongs to \(\mathcal {C}_{T_o}\). On this regard, see as well Proposition 3.1 below.
2.3 Weak solution of the limit equation
Recalling the definitions of \(p_\pm \) in (1.5), we introduce the bilinear form
and the associated quadratic form \(\hat{\mathcal E}[u]:=\hat{\mathcal E}[u,u]\). We now define the semi-norm \(\Vert u\Vert _{\mathcal {H}_o}\,:= \, \hat{\mathcal E}^{1/2}[u]\) for any Borel function \(u:{\mathbb {R}}_*\rightarrow {\mathbb {R}}\) such that the expression is finite. One can check that \(\hat{\mathcal E}^{1/2}[u]<+\infty \), if u belongs to \(C_c^\infty ({\mathbb {R}}_*)\). We then denote by \(\mathcal {H}_o\) the completion of \(C_c^\infty ({\mathbb {R}}_*)\) under \(\Vert \cdot \Vert _{\mathcal {H}_o}\). We proceed to the definition of a weak solution to (1.17) with the interface conditions (1.2).
Definition 2.4
A bounded function \(\bar{W}:\bar{{\mathbb {R}}}_+\times {\mathbb {R}}\rightarrow {\mathbb {R}}\) is a weak solution to (1.17) with interface conditions (1.2) if
-
(i)
\(\bar{W}(\cdot )-T_o\in L^2_{\text {loc}}([0,+\infty );\mathcal {H}_o)\) and \(\bar{W}(\cdot )-T_\infty \in C([0,+\infty );L^2({\mathbb {R}}))\) for some \(T_\infty \in {\mathbb {R}}\),
-
(ii)
for any \(F\in C^\infty _c([0,+\infty )\times {\mathbb {R}}_*)\) and any \(t\ge 0\), it holds that
$$\begin{aligned}{} & {} \int _{{\mathbb {R}}}F(0,y)[ \bar{W}_0(y)-T_o]\, dy =\int _{{\mathbb {R}}}F(t,y)[\bar{W}(t,y)-T_o]\, dy \nonumber \\{} & {} \quad -\int _0^{t}\int _{{\mathbb {R}}}\partial _sF(s,y) [\bar{W}(s,y)-T_o]\, dy ds +\bar{\gamma } \int _0^{t}\hat{\mathcal E} [F(s,\cdot ),\bar{W}(s,\cdot )-T_o] \, ds.\nonumber \\ \end{aligned}$$(2.7)
By Proposition 3.7 below, our definition guarantees the uniqueness of solutions.
2.4 Statement of the main result
Once we have formulated the definition of a solution, we are ready to formulate rigorously our main result. It reads as follows:
Theorem 2.5
Assume that the hypotheses about \(p_\pm \), \(p_0\), R and \(\bar{\omega }\) made between (1.8) and (1.15) hold. Furthermore, suppose that assumptions (2.3) are in force. Let \(T_\infty \in {\mathbb {R}}\) and \(W_0\in {\mathcal C}_{T_o}\) be such that \(\bar{W}_0-T_\infty \in L^2({\mathbb {R}})\) and \(\bar{W}_0-T_o\in \mathcal {H}_o\), with \(\bar{W}_0\) defined in (1.16). Let \(W_\lambda (t,y,k)\) be the solution, in the sense of Definition 2.3, to the Cauchy problem (1.6). Then,
for all \(t>0\) and test functions \(F\in C^\infty _c({\mathbb {R}}\times {\mathbb {T}})\). Moreover, the limit \(\bar{W}(t,y)\) is the weak solution, in the sense of Definition 2.4, to the Cauchy problem (1.17) with the initial distribution \(\bar{W}_0\) and the fractional diffusion coefficient
3 Probabilistic representation of solutions
3.1 Construction of the position-frequency process
As usual, \({\mathbb {N}}:=\{1,2,\ldots \}\) and \({\mathbb {N}}_0:=\{0,1,2,\ldots \}\). Let \((\Omega ,\mathcal {F},\mathbb {P})\) be a probability space carrying the following random objects. We consider a Markov chain \(K_n(k)\) and the renewal process \({\mathfrak T}_n(k)\). More precisely, \(\{K_n(k)\}_{n\in {\mathbb {N}}_0}\) is such that \(K_0(k)=k\) and \(\{K_n(k)\}_{n\in {\mathbb {N}}}\) are i.i.d. random variables on \({\mathbb {T}}\), distributed according to the following probability measure:
and
where \(\{\tau _n\}_{n\in {\mathbb {N}}_0}\) is an independent sequence of i.i.d. exponentially distributed random variables with intensity 1 and, we recall, \(\bar{t}(k)\) was defined in (2.1). We introduce the process \(\{{\mathfrak T}(t,k)\}_{t\ge 0}\) as the linear interpolation between the values of \({\mathfrak T}_n(k)\):
We then define the continuous-time frequency process as
Here \({\mathfrak T}^{-1}\) is the inverse function of \(t\mapsto {\mathfrak T}(t,k)\) and \([\cdot ]\) denotes the integer part.
According to (2.2), the particle position at the time of the n-th renewal of its frequency is
In particular, the law of \(Z_n(y,k)\), for each \(n\in {\mathbb {N}}\), is absolutely continuous with respect to the Lebesgue measure on \({\mathbb {R}}\). We then consider an auxiliary Poisson process \(\{N(t)\}_{t\ge 0}\) of intensity 1 that is independent of both \(\{K_n\}_{n\ge 0}\) and \(\{\tau _n\}_{n\ge 0}\). We define \(\{\tilde{N}(t)\}_{t\ge 0}\) as the linear interpolation between the nodal points of N. Namely, if \(n:=N(t)\) then
where \(l:=\inf \{s:N(s)=n\}\) and \(r:=\inf \{s: N(s)=n+1\}\). Our next process, \(\{\tilde{Z}(t,y,k)\}_{t\ge 0}\), is obtained by linearly interpolating between the nodal points of \(Z_{N(t)}(y,k)\): if \(n=N(t)\), then
Next we define the transmission-reflection-killing mechanism at the interface \(o=[y=0]\). To this end, we fix \((y,k)\in {\mathbb {R}}_*\times {\mathbb {T}}_*\) and consider the times \(\left\{ n_{m}\left( \Big \{Z_n(y,k)\Big \}_{n\ge 0}\right) \right\} _{m\ge 0}\) when \(\Big \{Z_n(y,k)\Big \}_{n\ge 0}\) crosses the interface. Namely, we let \(n_0:=0\) and then define recursively
For the sake of brevity, when there is no danger of confusion, we skip the sequence \(\Big \{Z_n(y,k)\Big \}_{n\ge 0}\) from the notation. Similarly, we define the sequence \(\Big \{\tilde{\mathfrak {s}}_{m}\Big (\{\tilde{Z}(t,y,k)\}_{t\ge 0} \Big )\Big \}_{m\ge 0}\) of consecutive times when the process \(\tilde{Z}(t,y,k)\) crosses the interface o. Namely, we let \(\tilde{\mathfrak {s}}_{0}=0\) and
Again, we simplify the notation by omitting the path of the process when there is no danger of confusion. Sometimes, to highlight the dependence of the crossing times on the starting point (which will be relevant in the argument), we may write \(\tilde{\mathfrak {s}}_{y,k,m}\), or \(\tilde{\mathfrak {s}}_{y,m}\). Let \(\mathfrak {s}_{m}:=\inf \{t>0:\,n_{m}=N(t)\}\). Since the m-th passage of \(\tilde{Z}(t,y,k)\) across the interface occurs before \(\mathfrak {s}_{m}\), we have \(\tilde{\mathfrak {s}}_{m} < \mathfrak {s}_{m}\), \(\mathbb {P}\)-a.s. Likewise, the \((m+1)\)-st passage has to occur after \(\mathfrak {s}_{m}\), so \(\mathfrak {s}_{m} < \tilde{\mathfrak {s}}_{m+1}\), \(\mathbb {P}\)-a.s.
We now consider a sequence \( \{\sigma _{m}\}_{m\ge 0}\) of \(\{-1,0,1\}\)-valued random variables that are independent when conditioned on \(\{K_n(k)\}_{n\ge 0}\), such that \(\sigma _0:=1\) and
Here and below, \(p_\iota \) means \(p_\pm \) if \(\iota =\pm 1\), respectively. Of course, \(\{\sigma _m\}_{m\ge 1}\) can be defined by applying the quantile functions for (3.5) to independent i.i.d. uniform random variables. We can finally add the random interface mechanism to the considered processes. Namely,
We define \({\mathcal S}(t,k):=\tilde{N}^{-1}\left( {\mathfrak T}^{-1}(t,k)\right) \) and \({\mathfrak {u}}_{m}:=\mathcal {S}^{-1}(\tilde{\mathfrak {s}}_{m},k)\). Let
The “true” position process is then given by
We now denote by \({\mathfrak f}:= \min \left\{ m\in {\mathbb {N}}:\sigma _{m}=0\right\} \) the interface crossing at which the particle gets killed and let \(\mathfrak { u}_{{\mathfrak f}}:={\mathcal S}^{-1}(\tilde{\mathfrak s}_{{\mathfrak f}},k)\). To highlight the dependence of the crossing times on the starting point, sometimes we may write \(\mathfrak { u}_{y,k,{\mathfrak f}}\). The following probabilistic representation of the solution to (1.1), with interface conditions (1.2), holds.
Proposition 3.1
Let \(W_0\) be in \(\mathcal {C}_{T_o}\). Then, for any \((t,y,k)\in [0,+\infty )\times {\mathbb {R}}_*\times {\mathbb {T}}_*\),
In addition, if \(W_0(0,k)=T_o\) for all \(k\in {\mathbb {T}}\), then
is the unique classical solution, in the sense of Definition 2.3, to the Cauchy problem (1.1) with initial distribution \(W_0\).
Proof
The existence and uniqueness of a classic solution and its representation by formula (3.8) have already been shown in [22], Section A, and Proposition 3.2. Recalling that \(Y^o( t,y,k)=0\) if \(t\ge {\mathfrak {u}}_{y,k,{\mathfrak f}}\), we can use that \(W_0(0,k)=T_o\) to get that
Then, (3.9) follows immediately from (3.8). \(\square \)
3.2 Scaling of the process \((Y^o( t,y,k), K^o( t,k))\)
Equations (1.6) may be considered a special case of (1.1), so we first focus on establishing a flexible notation for later use. As before, we fix \((y,k)\in {\mathbb {R}}_*\times {\mathbb {T}}_*\). We let \(\lambda >0\) and rescale the clock processes, introduced in the previous section, as follows:
and
We then notice that for large \(\lambda \), the process \({\mathcal S}_\lambda (t,k)\) becomes almost deterministic. More precisely, by (1.15) we have that
Except for the first deterministic term, the inverse \({\mathcal S}_\lambda ^{-1}(s,k) = {\mathfrak T}_\lambda (\tilde{N}_{\lambda }(s),k)\) can be represented as a Poisson sum of i.i.d. variables with finite first moment and so, by a standard argument using the strong law of large numbers, it can be shown that:
Proposition 3.2
For any \(t_*\in [0,+\infty )\), we have
We can now rescale the “position” process in the following way. Let us define \(\{\tilde{Z}_\lambda (t,y,k)\}_{t\ge 0}\) as the linear interpolation between the nodal points of \(Z_\lambda (t,y,k)\), where
We then construct the sequences \(\{n_{m}^\lambda \}_{m\ge 0}\), \(\{{{\mathfrak {s}}}_{m}^{\lambda }\}_{m\ge 0}\), \(\{{\tilde{\mathfrak {s}}}_{m}^{\lambda }\}_{m\ge 0}\) of indices and stopping times for the rescaled processes \(Z_\lambda (t,y,k)\), \(\tilde{Z}_\lambda (t,y,k)\), defined by analogues of (3.3) and (3.4), skipping (y, k) from the notation, when unambiguous. Since the law of \( Z_\lambda (t,y,k)\) is absolutely continuous with respect to the Lebesgue measure, for each \(y\in {\mathbb {R}}\) there exists a strictly increasing sequence \(n_{m}^\lambda :=N_\lambda ({{\mathfrak {s}}}_{m}^{\lambda })\) such that \( {{\mathfrak {s}}}_{m}^{\lambda } \, = \, \tilde{N}_\lambda ^{-1}\left( n_{m}^\lambda \right) . \) To set the notation for the processes subject to the random mechanism at the interface, let \(\{\sigma _{m}^\lambda \}_{m\in {\mathbb {N}}}\) be a sequence of \(\{-1,0,1\}\)-valued random variables that are independent when conditioned on \(\{K_n(k)\}_{n\ge 0}\) and such that
We then set
The jump process \(\{Z_\lambda ^o(t,y,k)\}_{t\ge 0}\) is now defined, for any \(m\in {\mathbb {N}}_0\), as
Here \(\sigma _{0}^{\lambda }:=1\). Similarly, the continuous trajectory process \(\{\tilde{Z}_\lambda ^o(t,y,k)\}_{t\ge 0}\) can be obtained as in (3.15) but with respect to the stopping times \(\{\tilde{\mathfrak {s}}^{\lambda }_m\}_{m\in {\mathbb {N}}_0}\).
In what follows, we show that it is unlikely that the scaled position process \(Z_\lambda (t,y,k)\) crosses the interface after the first jump, when \(\lambda \) becomes large. Indeed, recall from (3.14) that \(Z_1^\lambda (y,k)=y-\lambda ^{-1/\alpha }S(k)\tau _0\) where \(\tau _0\) is \(\exp (1)\)-distributed. Let
The following estimate follows immediately.
Proposition 3.3
For all \(\lambda >0\), \(y\in {\mathbb {R}}_*\) and \(k\in {\mathbb {T}}_*\) we have \({\mathbb {P}}\left( A^{\lambda }(y,k)\right) \, \le \, \exp \left\{ -\frac{|y|\lambda ^{1/\alpha }}{|S(k)|}\right\} .\)
3.3 Auxiliary stable Lévy process
To describe the limit, as \(\lambda \) goes to \(+\infty \), of processes \(\{Z_\lambda ^o(t)\}_{t\ge 0}\), we consider the \(\alpha \)-stable Lévy process \(\{\zeta (t)\}_{t\ge 0}\) with the infinitesimal generator
where \(\bar{r}_*\,:= \, R^*_2S_*^{1+\alpha } \Gamma (\alpha +1)/S'_*\). For any \(y\ne 0\), we denote \(\zeta (t,y):= y+\zeta (t)\). By a straightforward modification of (3.4), we let \(\mathfrak {t}_{0}:=0\) and denote by \(\Big \{\mathfrak {t}_{m}\Big (\Big \{\zeta (t,y)\Big \}_{t\ge 0}\Big )\Big \}_{m\ge 1}\) the consecutive times when the process \(\zeta (t,y)\) crosses the interface o. We often abbreviate
and we also let
It is known that \(\mathfrak {t}_{y,\mathfrak {f}}\) is finite \(\mathbb {P}\)-a.s. and its law is absolutely continuous with respect to the Lebesgue measure on \({\mathbb {R}}\) (cf. [30, Example 43.22]). We can finally introduce the tentative limit process \(\{\zeta ^o(t,y)\}_{t\ge 0}\) as follows. Recalling the definitions of \(p_{\pm }\) in (1.5) and \(p_0=0\), stemming from (1.8), we take a sequence \(\{\sigma _m\}_{m\in {\mathbb {N}}}\) of i.i.d. \(\{-1,1\}\)-valued random variables such that for any \(m\in {\mathbb {N}}\), the random variable \(\sigma _m\) is independent of \(\{\zeta (t,y)\}_{t\ge 0}\) and \(\mathbb {P}(\sigma _m=\pm 1) = p_\pm \) (the notation of (3.5) is best ignored from now on). Letting \(\sigma _0:=1\), we define for \(m\in {\mathbb {N}}_0\),
and \(\zeta ^o(t,y):= 0\) if \(t\ge \mathfrak {t}_{y,\mathfrak {f}}\). The process, discussed in detail below, is an interesting take on the question of constructing skew stable Lévy processes, a topic recently examined in [18]. The case of a diffusive limit, when the transmission probability depends on whether the transmission occurs from the left to the right half-line or otherwise, leads to the limit which is given by a skew Brownian motion, see [8, Theorem 3.1].
However, in view of limit theorems for kinetic equations, the following is our main motivation.
Theorem 3.4
When \(\lambda \) tends to \(+\infty \), the processes \(\left\{ Z_{\lambda }^o(t,y,k)\right\} _{t\ge 0}\) converge both in finite distributions and weakly, over \(\mathcal {D}[0,+\infty )\) with the \(J_1\)-topology, to \(\left\{ \zeta ^o(t,y)\right\} _{t\ge 0}\).
The proof of the theorem shall be presented in Sect. 5 below.
For simplicity, denote by \(\mathcal {X}\) the space \(\mathcal {D}[0,+\infty )\times \mathcal {C}[0,+\infty )\) equipped with the product of the \(M_1\)-topology and the topology of uniform convergence over compact intervals. The precise definition of the \(M_1\)-topology on \(\mathcal {D}[0,+\infty )\) can be found in [31, Section 12]. We only mention here that the topology is metrizable and for two paths to be close, it means that they have their parametrizations, both in the temporal and spatial domains, that are not far from each other in the supremum norm, see [31, formula (3.7), Section 12, p. 395].
It then follows from Theorem 3.4 and Proposition 3.2 that:
Corollary 3.5
As \(\lambda \rightarrow +\infty \), the processes \(\left\{ \left( \tilde{Z}_{\lambda }^o(t,y,k),\mathcal {S}_{\lambda }(t,k)\right) \right\} _{t\ge 0}\) converge both in finite distributions and weakly, over \(\mathcal {X}\), to \(\left\{ \left( \zeta ^o(t,y),\bar{\theta }t\right) \right\} _{t\ge 0}\).
Let us denote \(K_\lambda (t,k):=K(\lambda t,k)\) and \( \mathfrak {u}_{m}^\lambda :={\mathcal S}_{\lambda }(\tilde{\mathfrak {s}}_{m}^\lambda ,k) \). We finally define the position- momentum process \(\Big \{K_\lambda ^o(t,k), Y_\lambda ^o(t,k,y)\Big \}_{t\ge 0}\) as
If we let \(\eta ^o(t,y):= \zeta ^o\left( \bar{\theta }t,y\right) \), where \(\bar{\theta }\) is defined in (3.11), then the following result is an immediate consequence of Corollary 3.5.
Corollary 3.6
As \(\lambda \) goes to \(+\infty \), the processes \(\left\{ Y^o_{\lambda }(t,y,k)\right\} _{t\ge 0}\) converge both in finite distributions and weakly, over \(\mathcal {D}[0,+\infty )\) with the \(M_1\)-topology, to the process \(\left\{ \eta ^o(t,y)\right\} _{t\ge 0}\).
Proof
Invoking Theorem 7.2.3 of [32] and using formula (3.21), we conclude the weak convergence of the processes. Since, for any deterministic time \(t\ge 0\) we have
by virtue of Lemma 6.5.1 in [31], the set of discontinuities of the one-dimensional projection mapping \(\omega \mapsto \eta ^o(t,y,\omega )\) is of null probability. Using the continuous mapping theorem, see Theorem 2.7 of [7], we conclude the convergence of the one-dimensional distributions. The generalisation to finite-dimensional distributions is trivial. \(\square \)
We conclude this section by presenting a probabilistic characterisation of the weak solution of the limit Cauchy problem (1.17) in terms of the process \(\eta ^o(t,y)\). A proof of this result can be found in the “Appendix B” below.
Proposition 3.7
Let \(\bar{W}_0-T_o\) be in \(\mathcal {H}_0\) such that \(\bar{W}_0-T_\infty \) is in \(L^2({\mathbb {R}})\) for some \(T_\infty \in {\mathbb {R}}\). Then, the function
is the unique weak solution, in the sense of Definition 2.4, to the Cauchy problem (1.17) with initial condition \(\bar{W}_0\).
4 Proof of Theorem 2.5
Since \(W_{\lambda }(t,y,k)-T_o\) is the solution of (1.6) with the interface condition (1.2) corresponding to the zero thermostat temperature, we assume without loss of generality that \(T_o=0\). Clearly, the solution \(W_\lambda \) is given by (3.8). In what follows, we often write \(W_\lambda (t)\) for \(W_\lambda (t,\cdot ,\cdot )\).
Fix a test function F in \(C^\infty _c({\mathbb {R}}\times {\mathbb {T}})\). Suppose that \(\textrm{supp}\,F\subset [-M_0,M_0]\times {\mathbb {T}}\). Thanks to Corollary 3.6 for any \(\varepsilon >0\) we can find a sufficiently large \(M>1\) such that
By a standard approximation argument, it is enough therefore to consider an initial condition \(W_0\in C^\infty _c({\mathbb {R}}\times {\mathbb {T}})\cap \mathcal {C}_{0}\). If the initial condition \(W_0\) is independent of k, i.e. \(W_0(y,k)=W_0(y)\), then, by Remark 2.2, we have \(W_0(0^+)=W_0(0^-)=0\) and we can use formula (3.9) to represent \(W_\lambda \). As a result, we can immediately conclude the proof of Theorem 2.5 using Proposition 3.1 (with respect to the rescaled processes), (3.22) and Corollary 3.6.
The next result enables to replace an arbitrary initial condition \(W_0(y,k)\) by its average over \({\mathbb {T}}\) with respect to the measure \(\pi (dk)=R_2(k)/\big ({\mathcal R}R_1(k)\big )dk\). For notational simplicity, we denote by \(L^2_{\pi }({\mathbb {R}}\times {\mathbb {T}})\) the \(L^2\)-space on \({\mathbb {R}}\times {\mathbb {T}}\) with respect to the product measure \(dy\, \pi (dk)\).
Lemma 4.1
Suppose that \(W_0\in C_c({\mathbb {R}}\times {\mathbb {T}}) \). Let \(\bar{W}_0\) be the function defined by (1.16). Then, for any \(\varepsilon >0\), there exists \({\delta }_0>0\) such that
The above lemma shall be proved at the end of this section.
Since \(W_\lambda (t)\) solves (1.6) in the sense of Definition 2.3, the existence of the directional derivative \(D_tW_{\lambda }(t,y,k)\), see (2.5), implies the differentiability of \(t\mapsto \Vert W_{\lambda }(t)\Vert ^2_{L^2_\pi ({\mathbb {R}}\times {\mathbb {T}})}\). Computing \(\frac{d}{dt} \Vert W_{\lambda }(t)\Vert ^2_{L^2_\pi ({\mathbb {R}}\times {\mathbb {T}})}\) and using (1.6) together with the interface conditions (1.2), see the calculations in [5, Section 3], we immediately conclude the following.
Proposition 4.2
The norm \( \Vert W_{\lambda }(t)\Vert _{L^2_\pi ({\mathbb {R}}\times {\mathbb {T}})}\), defined as a function of \(t\in [0,\infty )\), does not increase.
We can now finish the proof of Theorem 2.5. We first show the following variant of (2.8),
Choose an arbitrary \(\varepsilon >0\). Let \(\delta >0\) be as in Lemma 4.1 and \(\tilde{W}_\lambda \) the solution to the Cauchy problem (1.6) when the initial condition is given at time \(t=\delta \) by \(\tilde{W}_\lambda (\delta ,y,k)=\bar{W}_0(y)\). In particular, we can use Proposition 3.1 to represent \(\tilde{W}_\lambda \) as
By virtue of Proposition 4.2 and Lemma 4.1, we have that
Therefore, it follows that
Recalling (3.22) and (4.3), it is not difficult to conclude from Corollary 3.6 that
Choosing \(\delta >0\) sufficiently small, we can then guarantee that
As a result, we conclude that for any \(\varepsilon >0\),
and (4.2) immediately follows. The conclusion of Theorem 2.5 with respect to the Lebesgue measure on \({\mathbb {T}}\), as in (2.8), immediately follows from (4.2) and the fact that \(\Vert F \tilde{W}_\lambda (t )\Vert _\infty \le \Vert F W_0\Vert _\infty \). The claim that \(\bar{W}(t,y)\) is a weak solution of (1.17) follows from Proposition 3.7.\(\square \)
4.1 Proof of Lemma 4.1
Recall that we have assumed that \(W_0\in C_c({\mathbb {R}}\times {\mathbb {T}}) \). Fix an arbitrary \(\varepsilon \in (0,1)\). We shall show that there exists \(\delta _0\in (0,1)\) such that
where \( \hat{W}^o_\lambda (t,y,k):= {\mathbb E}\left[ W_0\left( y, K_\lambda ^o(t,k)\right) \right] \). Indeed, let \(\rho >0\) be so small that
for all \(\delta >0\). This is possible because the volume of integration can be made small and \(W_{\lambda }(\delta )\) is bounded in the \(L^\infty \)-norm (thanks to (3.8)).
Since \(W_0\) is compactly supported we claim that we can choose \(\rho \in (0,1)\) so small that
for any \(\delta >0\). Indeed, for a small \(\rho \) we have \(\hat{W}^o_\lambda (t,y,k)=0\) for \(|y|\ge \rho ^{-1}\) and all \((t,k)\in \bar{\mathbb {R}}_+\times {\mathbb {T}}_*\). To conclude (4.6) it suffices to show that
The function
is continuous, compactly supported and even. Since \(|Y^o_\lambda (\delta ,y,k)|=|Y_\lambda (\delta ,y,k)|\) for \(\delta \in [0, \mathfrak { u}_{y,k,{\mathfrak f}}^{\lambda })\) we have \(W_0^*\left( Y^o_\lambda (\delta ,y,k)\right) = W_0^*\left( Y_\lambda (\delta ,y,k)\right) \) for \(\delta \in [0, \mathfrak { u}_{y,k,{\mathfrak f}}^{\lambda })\). As a result, thanks to (3.8) with \(T_o=0\),
Using the weak convergence of \(\Big \{Y_\lambda ( t,0,k)\Big \}_{t\ge 0}\) to \(\Big \{\eta (t)\Big \}_{t\ge 0}\) we can write
Estimate (4.7) follows taking first the supremum over \(\delta \in (0,1]\) and then the limit as \(\rho \) goes to zero.
To conclude the proof of (4.4) we need to prove yet that given \(\rho \in (0,1)\), there exist \(\delta _0\in (0,1)\) such that
for \(\delta \in (0,\delta _0].\)
We start with the following observation. For any \(\rho '\in (0,\rho )\) and \(k\in {\mathbb {T}}_*\) we have
where \(Y_\lambda (t,y,k)\) is the analogue of \(Y_\lambda ^o( t,y,k)\), but without the interface (or, equivalently, with \(p_+\equiv 1\) in our model). Taking the complements in (4.9) and using \(Y_\lambda (t,y,k)-y=Y_\lambda (t,0,k)\), we conclude that
The processes \(\Big \{Y_\lambda ( t,0,k)\Big \}_{t\ge 0}\) converge, as \(\lambda \rightarrow +\infty \), both in finite distributions and weakly, over \(\mathcal {D}[0,+\infty )\) with the \(M_1\)-topology, to a symmetric, \((1+1/\beta )\)-stable process \(\Big \{\eta (t)\Big \}_{t\ge 0}\), see [4, Theorem 3.2], or [20, Theorem 3.1]. Therefore by (4.10) and [32, Theorem 7.4.1] we conclude that for any \(k\in {\mathbb {T}}\) and \(\rho '< \rho \), there exists a sufficiently small \(\delta _0>0\) such that
for all \(\delta \in (0,\delta _0]\). In particular, it follows that
where \(\mathfrak {u}^\lambda _{y,k,1}\), \(\mathfrak {u}^\lambda _{y,k,\mathfrak {f}}\) are the first time the process \(Y_\lambda (t,y,k)\) crosses the interface, and the time it gets killed, correspondingly. Coming back to the proof of (4.8), note that, by virtue of (3.8) (with \(T_o=0\)), the expression inside the integral there can be rewritten as
Using uniform continuity of \(W_0\) and (4.9) we can select \(\rho '\) in such a way that for some \(\delta _0\in (0,1)\), we have
provided \(\delta \in (0,\delta _0]\). Combining this with (4.12), we conclude that for each \(k\in {\mathbb {T}}_*\), there exists \(\delta _0\in (0,1)\) such that
for \(\delta \in (0,\delta _0]\). Since the expression inside the integral in (4.8) is uniformly bounded in \(\lambda \), we can use the ”limsup version” of the Fatou lemma to conclude that (4.8) holds. This ends the proof of (4.4).
4.1.1 The end of the proof of Lemma 4.1
Let \( \hat{W}_\lambda (t,y,k):={\mathbb E}\left[ W_0\left( y,K_\lambda (t,k) \right) \right] \). Using (4.12) and the fact that \(W_0\) is compactly supported, we conclude that for any \(\varepsilon >0\), there exists \(\delta _0>0\) such that
for all \(\delta \in (0,\delta _0]\). The dynamics of the frequency process \(K_\lambda ( t,k)\) is reversible with respect to the measure \(\pi \) on the torus \({\mathbb {T}}\) and 0 is a simple eigenvalue for the generator \(L_k\), i.e. \(L_ku=0\) implies \(u\equiv \textrm{const}\), \(\pi \) a.e. The latter can be easily seen, due to the fact that \(L_k1=0\) and the identity
which holds for all \(u\in L^2(\pi )\). As a consequence, we have
for all \(\delta \in (0,\delta _0]\). Combining with (4.4), this ends the proof of the lemma. \(\square \)
5 Proof of Theorem 3.4
To prove Theorem 3.4, we show the convergence of the Markov semigroups corresponding to processes \(\left\{ Z_{\lambda }^o(t,y,k)\right\} _{t\ge 0}\), see (3.15). In the first part of the present section, we focus on constructing a Lévy-type process \(\{\hat{Z}^o_\lambda (t,y)\}_{t\ge 0}\) whose increments, after the first jump, coincide with those of \(\left\{ Z_{\lambda }^o(t,y,k)\right\} _{t\ge 0}\).
5.1 Construction of the associated Markov process
Let us denote by r(y) the density of \(S(K_1)\), where S(k) is defined in (2.2) and \(K_1\) is distributed according to \(R_2(k)dk\) (see (3.1)). Recalling that the function S(k) is bijective, the law of \(S(K_1)\) is then given by
Since S is odd, its law \(\mu _S\) is even, and thus r(y) is even. Let us now define, for simplicity,
for any \(\lambda >0\) and any z in \({\mathbb {R}}\), with the classical notation \(\tilde{p}_{\iota }(z):=\tilde{p}_{\iota ,1}(z)\). Clearly, all the above functions are even. Thanks to (1.8) we may assume, with no loss of generality, that
From the general assumptions of the model (cf. (1.8), (1.11), (1.12) and (2.3)), it is not difficult to verify, by direct calculations, the following properties:
Lemma 5.1
We have
and
In addition,
where \( \tilde{p}_*:= p_*\beta _3^{\kappa }, \) and \(p_*\) is given by (1.8).
Consider, for any y in \({\mathbb {R}}\),
and its rescaled version \(\bar{r}_\lambda (y):=\lambda ^{1+1/\alpha } \bar{r}\left( \lambda ^{1/\alpha }y\right) \) for any \(\lambda >0\). Lemma 5.1 then implies that
We now consider the Markov process \(\left\{ \hat{Z}_\lambda ^o(t,y)\right\} _{t\ge 0}\) starting at y, whose generator is defined on \(C_0({\mathbb {R}})\) - the space of continuous functions vanishing at \(\infty \) - by \({\hat{\mathcal {L}}}_{\lambda }^o u(0)=0\) and, for any \(y\in {\mathbb {R}}_*\), by
where the jump and killing kernels are given by
We will prove at the end of the present section that the Markov processes \(\hat{Z}^o_\lambda (t,y)\) converge as well to the limit process \(\zeta ^o(t,y)\), defined in (3.19).
Theorem 5.2
Let y be in \({\mathbb {R}}_*\). As \(\lambda \rightarrow +\infty \), the processes \(\{\hat{Z}_{\lambda }^o(t,y)\}_{t\ge 0}\) converge both in finite distributions and weakly, over \(\mathcal {D}[0,+\infty )\) with the \(J_1\)-topology, to \(\left\{ \zeta ^o(t,y)\right\} _{t\ge 0}\).
5.2 Properties of the Markov semigroup corresponding to \(\{\hat{Z}^o_\lambda (t,y)\}_{t\ge 0}\)
In order to show Theorem 5.2, we will strongly rely on a convergence property between the corresponding Markov semigroups. The process \(\hat{Z}_\lambda ^o(t,y)\) killed at the interface is Markovian and its transition semigroup is given by
Here \(\hat{\mathfrak s}^{\lambda }_{y,{\mathfrak f}}\,:= \, \inf \{t>0:\hat{Z}_\lambda ^o(t,y) =0\}\). According to [15, Corollary 4.2.8, p. 170], \(\Big (P_t^{o,\lambda }\Big )\) forms a strongly continuous semigroup of contractions on \(C_0({\mathbb {R}})\).
We shall also consider the process stopped at \(\hat{\mathfrak s}^{\lambda }_{y,{\mathfrak f}}\). Its transition semigroup equals
According to Theorem 5.2 the process \(\zeta ^o(t,y)\) is the limit of the killed processes \(\{\hat{Z}^o_\lambda (t,y)\}_{t\ge 0}\). Its Markov semigroup satisfies the following.
Proposition 5.3
For any y in \({\mathbb {R}}_*\), the process \(\{\zeta ^o(t,y)\}_{t\ge 0}\) generates a symmetric, strongly continuous Markov semigroup \(\{P_t^o\}_{t\ge 0}\) on \(L^2({\mathbb {R}})\) given by
Proof
We firstly note that (5.13) is well defined on \(L^2({\mathbb {R}})\). Let \(u_*(y):=|u(y)|+|u(-y)|\). From (3.19), we see that \(|\zeta ^o(t,y)|=|\zeta (t,y)|\), for \(t< \mathfrak {t}_{y,\mathfrak {f}}\). Therefore, since \(u_*\) is even, we can write
where \(\{P_t\}\) is the transition semigroup corresponding to the stable process \(\{\zeta (t,y)\}_{t\ge 0}\) - well defined on any \(L^p({\mathbb {R}})\), see, e.g., [2, Section 3.4]. \(\square \)
5.2.1 Proof that \((P^o_t)_{t\ge 0}\) is a Markovian semigroup
Fix \(t,s>0\) and \(N\in {\mathbb {N}}\). Let us consider \(0<s_1<\dots <s_N\le s\), a finite family \(\{\phi _j, \, j=1,\ldots ,N \}\) of bounded Borel measurable functions \(\phi _j:{\mathbb {R}}_*\rightarrow {\mathbb {R}}_+\) and a bounded Borel measurable function \(u:{\mathbb {R}}_*\rightarrow {\mathbb {R}}_+\). It is then enough to show that
where we denoted, for simplicity, \(\Phi (y)=\prod _{j=1}^N\phi _j(\zeta ^o(s_j,y))\). Recalling the definition of \(\zeta ^o(t,y)\) in (3.19), we rewrite the left-hand side of (5.14) as:
Let now \(\{ \tilde{\zeta }(t)\}_{t\ge 0}\) be an independent copy of the stable process \(\zeta (t)\). Similarly to (3.4), we can also consider, for any \(m\in \{0,1,\ldots \}\) and \(z\in {\mathbb {R}}_*\), the m-th consecutive time \(\tilde{\mathfrak t}_{z,m}\) that the process \(\{z+\tilde{\zeta }(t) \}_{t\ge 0}\) crosses the interface. Using the independence of increments for the stable Lévy process, we then rewrite the right-hand side of (5.15) as
Then, using the symmetry of the law of a stable process, it follows that for any \(\varepsilon =\pm 1\) and any z in \({\mathbb {R}}_*\), we have that
Therefore, we can finally rewrite (5.16) as:
and (5.14) then follows immediately.
The semigroup \(P^o_t\) is symmetric. Fixed u, v in \(L^2({\mathbb {R}})\), our aim is to show the following:
We start by rewriting the left-hand side of (5.17) as
We have used the convention \(p_{\pm 1}:= p_{\pm }\). By a standard argument, using the Chapman-Kolmogorov equations, it is possible to show the following.
Lemma 5.4
Let \(F:\mathcal {D}[0,t]\rightarrow {\mathbb {R}}_+\) be bounded measurable. Then,
where the reversed time process is given by \(\zeta ^r(s,y):=\zeta ((t-s)_{-},y)\), \(s\in [0,t]\).
We can now exploit the above lemma with \(F(\zeta )=\mathbbm {1}_{\mathfrak {t}_{m}(\zeta ) \le t< \mathfrak { t}_{m+1}(\zeta )}\). Note that \(F(\zeta )=F(\zeta ^r)\). Therefore, we conclude that
If we assume for the moment that \(\prod _{j=1}^m\varepsilon _j=1\), it then obviously follows from (5.20) that
where, we recall, \({\mathfrak t}_{y,m}\) is defined in (3.18). Suppose now that \(\prod _{j=1}^m\varepsilon _j=-1\). Since the laws of \(\{\zeta (t,y)\}_{t\ge 0}\) and \(\{-\zeta (t,-y)\}_{t\ge 0}\) are identical, we can rewrite (5.20) as
Applying (5.21)–(5.22) to (5.18) above, we can finally conclude the proof of (5.17). From the latter we infer that each \(P_t^o\) is a contraction in any \(L^p({\mathbb {R}})\), \(p\in [1,+\infty )\). It is easy to check that \(\lim _{t\rightarrow 0+}P_t^of=f\) in the \(L^2\)-sense for any compactly supported continuous f. Since the set of such functions is \(L^2\)-dense, the strong continuity of the semigroup follows. \(\square \)
As in (5.12), it is now clear that for any \(u\in C_0({\mathbb {R}})\) we have
where \(\hat{P}_{t}^{o}u(y)\) is the semigroup associated with \(\{\zeta ^o(t,y)\}_{t\ge 0}\). We will show in Sect. 6 below that the following result holds:
Theorem 5.5
As \(\lambda \) tends to \(+\infty \), the semigroups \(\{ P_{t}^{o,\lambda }\}_{t\ge 0}\), defined in (5.11), strongly converge in \(L^2({\mathbb {R}})\), uniformly on compact intervals, to the semigroup \(\{P_{t}^o\}_{t\ge 0}\) given in (5.13).
Thanks to the above result, we can now prove that the Markov process \(\hat{Z}_{\lambda }^o(t,y)\) converges to \(\zeta ^o(t,y)\).
5.3 Proof of Theorem 5.2
5.3.1 Convergence of finite dimensional distributions
Since the generalisation to finite distribution marginals is immediate, we will show only the convergence of the one-dimensional distributions. Recalling the definition of the process \(\{\hat{Z}^o_\lambda (t,y)\}_{t\ge 0}\) in (5.10), we start by considering the symmetric Lévy process \(\{\hat{Z}_{\lambda }(t)\}_{t\ge 0}\) corresponding to the following characteristic function:
where the Lévy symbol \( \Psi _\lambda (\xi ):=-\log \mathbb {E}\left[ e^{i\xi \cdot \hat{Z}_{\lambda }(1)}\right] \) can be written, thanks to (5.3), as
As before, we then denote \(\hat{Z}_\lambda (t,y):=y+\hat{Z}_\lambda (t)\) for any \(y\in {\mathbb {R}}^*\). If we assume for the moment that \(|\xi | \ge \lambda ^{1/\alpha }\), we have that
where
On the other hand, if \( |\xi | \le \lambda ^{1/\alpha }\), we can write from (5.24) that
Fixing \(c>0\), we now notice from the above controls that \(\Psi _\lambda (\xi )\succeq \theta _*>0\) for all \(|\xi |\ge c\). Then, it follows that the random variable \(\hat{Z}_\lambda (t)\) admits a density \(f_\lambda (t,\cdot )\) and belongs to \( L^1({\mathbb {R}})\cap L^\infty ({\mathbb {R}})\). Noticing that the process \(\{\hat{Z}_{\lambda }^o(t,y)\}_{t\ge 0}\) can be obtained from \(\{\hat{Z}_{\lambda }(t,y)\}_{t\ge 0}\) by performing a transformation of the trajectory (consisting in transmission-reflection, or moving the particle to the interface) described in Sect. 3.2, we have in particular that
Since we already know the weak convergence of the laws of \(\Big (\hat{Z}_{\lambda }(t,y)\Big )\), it in turn implies the tightness of the laws of the random variables \(\hat{Z}_{\lambda }^o(t,y)\), as \(\lambda \rightarrow +\infty \). To conclude the proof, it is enough to show that for any \(y\in {\mathbb {R}}_*\) and any \(\phi \in C^\infty _c({\mathbb {R}})\), we have
Let \(\varepsilon >0\) be arbitrary. Choosing \(h\in (0,t]\) sufficiently small, we can guarantee that
where \({\mathfrak t}_{y,1}\), \(\hat{\mathfrak {s}}^\lambda _{y,1}\) represents the first time the processes crosses the interface o (cf. (3.4)). Recalling the definition of the semigroups \(\hat{P}^{o,\lambda }_t\), \(\hat{P}^{o}_t\) in (5.12) and (5.13), we then have that
By the above reasoning and (5.12), it now follows that
Thanks to Proposition 5.5 and [17, Theorem 46.2], we know that \( P^{o,\lambda }_{t-h}\phi \) and \(f_{\lambda }(h,\cdot )\) strongly converge in \(L^2({\mathbb {R}})\), as \(\lambda \rightarrow +\infty \), to \( P^o_{t-h}\phi \) and \(f(h,\cdot )\), the density of \(\zeta (h)\), respectively. Therefore,
To deal with the above limit we use the following lemma, whose proof is presented in Sect. 5.3.2.
Lemma 5.6
Let f be in \(L^1({\mathbb {R}})\). Then,
Recalling that \(f_{\lambda }(h,\cdot )\) converge to \(f(h,\cdot )\) in \(L^2({\mathbb {R}})\), Lemma 5.6 implies that
Hence, from (5.32), (5.33), (5.35), and (5.23), we conclude that
Combining the above equation with (5.30) and (5.31), we can easily conclude the proof of the convergence of finite dimensional distributions.
5.3.2 Proof of Lemma 5.6
We are going to show (5.34) only for f in \(C^\infty _c({\mathbb {R}})\). The general statement of Lemma 5.6 then follows from a density argument. Fixed \(M>0\), let us consider an even, smooth function \(\phi _M:{\mathbb {R}}\rightarrow [0,1]\) such that
We can then define \(\psi _M:=1-\phi _M\). Using (5.28), we get that
for any \(|y|\le M/2\). Clearly, a similar reasoning holds for \(\zeta ^o(t,y)\) as well. Therefore, for any \(\varepsilon >0\), we can choose \(M>0\) and \(\lambda _0>1\) large enough, so that for \(\lambda \ge \lambda _0\), \(|y|\le M/2\), we have
Using again that M is large enough so that \(\text {supp} f\subseteq [-M/2,M/2]\), we then conclude that
The above estimate and Proposition 5.5 finally imply that
and (5.34) immediately follows. \(\square \)
5.3.3 Weak convergence in \(J_1\)-topology
By the previous argument, it is enough to prove tightness of the laws of \(\{\hat{Z}_{\lambda }^o(t,y)\}_{t\ge 0}\) over \(\mathcal {D}[0,+\infty )\) with the \(J_1\)-topology. We shall make use of the following notation. Given a function f in \(D[0,t_*]\), \(t_*>0\) and \(a<b\) in \([0,t_*]\), we denote \( \omega \left( a,b,f\right) := \sup \{|f(t)-f(s)|:a \le s<t<b\}\). We can then define the D-modulus of f of step \(\delta >0\) as:
where \(\mathbb {I}_\delta \) is composed of all the partitions \(\mathcal {P}=\{t_0=t<\dots <t_N=t_*\}\) of \([0,t_*]\) such that \(t_j-t_{j-1}\ge \delta \), for any \(j=1,\ldots ,N\). It is not difficult to check that the laws of \(\{\hat{Z}_\lambda (t,y)\}_{t\ge 0}\) are tight over \(\mathcal {D}[0,+\infty )\) with the \(J_1\)-topology. Theorem 13.2 in [7] then implies that for any \(t_*>0\) and any \(\varepsilon >0\), it holds that
Moreover, using nested partitions of the time interval \([0,t_*]\), one can easily show that
Thanks to the above control and (5.28), we therefore conclude that (5.38) and (5.39) hold for \(\hat{Z}_\lambda ^o(t,y)\) as well and the tightness of the latter, as \(\lambda \rightarrow +\infty \), immediately follows. \(\square \)
5.4 The end of the proof of Theorem 3.4
The conclusion of Theorem 3.4 follows from Theorem 5.2 and Proposition 3.3. Indeed, suppose that \(\{\hat{Z}_{\lambda }^o(t,y)\}_{t\ge 0}\), \(y\in {\mathbb {R}}_*\) is the process defined in the previous section. Let \(\tau _0\) be an exp(1) distributed random variable independent of the process. Define also the random variable \(\sigma \) that is independent of the process and \(\tau _0\) such that \({\mathbb {P}}[\sigma =\iota ]=p_\iota (k)\), \(\iota \in \{-1,0,1\}\). Recall that \(A^\lambda (y,k)\) is defined by (3.16) and \(Z_1^\lambda (y,k)=y-\lambda ^{-1/\alpha }S(k)\tau _0\). Let \(A^\lambda _\iota (y,k):=\{yZ_1^\lambda (y,k)<0\text { and } \sigma ^{\lambda }=\iota \}.\) Note that
Convergence of finite dimensional distributions, claimed in Theorem 3.4, is then a consequence of Proposition 3.3 and Theorem 5.5. Tightness can be argued from tightness of \(\Big (Z_{\lambda }(t,y,k)\Big )\) analogously to Sect. 5.3.3. \(\square \)
6 Proof of Theorem 5.5
We are going to show here the strong \(L^2\)-convergence of the semigroups \( P_{t}^{o,\lambda }\), defined in (5.11) and associated with the Markov processes \(\hat{Z^o}_\lambda (t,y)\), to the semigroup \(\{P_{t}^o\}_{t\ge 0}\), given in (5.13) and corresponding to the process \(\zeta ^o(t,y)\). The main tool used in the proof is the notion of convergence of the Dirichlet forms corresponding to the semigroups in the sense of Mosco. Before going into the actual proof, we recall some basic facts on the subject.
6.1 Basic notions on Sobolev spaces and Dirichlet forms
For any \(b\in (0,2)\), we define the following quadratic form:
for any Borel function \(u:{\mathbb {R}}\rightarrow {\mathbb {R}}\). We admit the possibility that the right-hand side of (6.1) equals infinity. Then, \(H^{b/2}({\mathbb {R}}):= \left\{ u\in L^2({\mathbb {R}}):{\mathcal E}[u] < \infty \right\} \) is a Hilbert space when endowed with the following norm
Recalling that \(\alpha \), defined in (1.14), belongs to (1, 2), we have that \(H^{\alpha /2}({\mathbb {R}})\subset C_b({\mathbb {R}})\) (see, e.g., [16, formula (1.4.33)]) and thus, u(0) is well defined. Moreover, \(C_c^\infty ({\mathbb {R}})\) is dense in \(H^{\alpha /2}({\mathbb {R}})\) and therefore (see [16, Example 1.4.1]) the form \({\mathcal E}\) is regular on \(H^{\alpha /2}({\mathbb {R}})\), i.e. there exists a set of compactly supported, smooth functions that is dense in \(H^{\alpha /2}({\mathbb {R}})\) (in the \(H^{\alpha /2}({\mathbb {R}})\)-norm), see [16, p. 6].
Recalling that \({\mathbb {R}}_*:={\mathbb {R}}\smallsetminus \{0\}\), we also consider the set \(H_0^{b/2}({\mathbb {R}}_*)\) obtained by the completion of \(C_c^\infty ({\mathbb {R}}_*)\) under the norm \(\Vert \cdot \Vert _{H^{b/2}({\mathbb {R}})}\). In particular, the space \(H_0^{\alpha /2}({\mathbb {R}}_*)\) can be equivalently characterised as the set of functions u in \(H^{\alpha /2}({\mathbb {R}})\) such that \(u(0)=0\). For a proof of this fact, see Corollary 6.4 below. It easily follows that the form \({\mathcal E}\) is regular on \(H_0^{\alpha /2}({\mathbb {R}}_*)\). Recall that \(\hat{\mathcal E}\) is the form defined in (2.6). Using (1.9), we easily conclude \(c{\mathcal E}\le \hat{\mathcal E}\), for some \(c\in (0,1)\). On the other hand, thanks to the Hardy inequality, see [14, Theorem 1.1, (T4)] and also Proposition 6.2 below, we can choose c so small that \(\hat{\mathcal E}\le c^{-1}{\mathcal E}\) on the space \(H^{\alpha /2}_0({\mathbb {R}}_*)\). Hence, the form \(\hat{\mathcal E}\) is comparable with \({\mathcal E}\) on the space \(H^{\alpha /2}_0({\mathbb {R}}_*)\). Thus, \(\hat{\mathcal E}\) is regular on \(H^{\alpha /2}_0({\mathbb {R}}_*)\).
To prove Proposition 5.5, we will show that the corresponding Dirichlet forms, defined below, converge in a suitable sense. More precisely (see [27, Definition 2.1.1]):
Definition 6.1
Let \(\{\mathcal {E}_{\lambda }\}_{\lambda >0}\) be a family of Dirichlet forms on \(L^2({\mathbb {R}})\) endowed with their natural domains:
Then, \(\mathcal {E}_\lambda \) is called M-convergent to a Dirichlet form \(\mathcal {E}_\infty \), as \(\lambda \rightarrow +\infty \), if for any \(u\in L^2({\mathbb {R}})\) the following conditions are satisfied:
-
i)
for any family \(\{u_{\lambda }\}_{\lambda >0}\) weakly convergent to u in \(L^2({\mathbb {R}})\), it holds that
$$\begin{aligned} \liminf _{\lambda \rightarrow +\infty }{\mathcal E}_{\lambda }[u_{\lambda }]\, \ge \, {\mathcal E}_\infty [u] \end{aligned}$$ -
ii)
there exists a family \(\{v_{\lambda }\}_{\lambda >0}\) strongly convergent to u in \(L^2({\mathbb {R}})\) such that
$$\begin{aligned} \limsup _{\lambda \rightarrow +\infty }{\mathcal E}_{\lambda }[v_{\lambda }]\, \le \, {\mathcal E}_\infty [u]. \end{aligned}$$
The notion of M-convergence is particularly useful for our purposes since it naturally implies (cf. [27, Corollary 2.6.1]) the strong convergence of the corresponding semigroups on \(L^2({\mathbb {R}})\).
We conclude this subsection presenting some properties of the Sobolev spaces we have just introduced. The proofs of the results formulated below can be found in “Appendix A”. Let us start with the following variant of the Hardy-type inequality of Dyda [14].
Proposition 6.2
For \(b\not =1\), there exists a positive constant \(C_b:=C(b)\) such that
Moreover, if \(1<b_0<b_1 <2\), then \(\sup _{b\in [b_0,b_1]} C_b \, < \, +\infty \).
We then obtain a characterisation of \(H_0^{b/2}({\mathbb {R}}_*)\) as a subspace of \(H^{b/2}({\mathbb {R}})\).
Proposition 6.3
For \(b\not =1\), we have
Finally, two other characterisations of the space \(H^{b_0/2}_0({\mathbb {R}}_*)\) are proposed for indices \(b_0\in (1,2)\).
Corollary 6.4
For every \(b_0\) in (1, 2), we have
6.2 Properties of the associated Dirichlet forms
Let us consider the Dirichlet form corresponding to \(P_t^{o,\lambda }\):
where the two kernels \(\hat{r}_{\lambda }\), \(k_{\lambda }\) were defined in (5.10). Note that these forms are finite for all \(u\in L^2({\mathbb {R}})\).
It follows from Proposition 5.3 that the process \(\{\zeta ^0(t,y)\}_{t\ge 0}\) admits a Dirichlet form \(\mathcal {E}^{o}\) given by
for any function u belonging to its natural domain \(\mathcal {D}(\mathcal {E}^o):=\{u\in L^2({\mathbb {R}}):\mathcal {E}^o[u]<\infty \}\). We extend \(\mathcal {E}^o\) to the entire \(L^2({\mathbb {R}})\) by letting \(\mathcal {E}^o[u]=+\infty \) for \(u\not \in \mathcal {D}(\mathcal {E}^o)\). Thanks to (6.10) below, the form \(\mathcal {E}^o\) is regular in the sense of [16, p. 6], i.e., \(C_c^\infty ({\mathbb {R}}_*)\) is its core. The main result of the present section is the following:
Theorem 6.5
The Dirichlet forms \(\{\hat{\mathcal E}_{\lambda }\}_{\lambda >0}\) are M-convergent, as \(\lambda \) goes to \(+\infty \), to the Dirichlet form \({\mathcal E}^o\).
Before proving Theorem 6.5, we show that \(\mathcal {E}^o\) actually coincides with the Dirichlet form \( \bar{r}_*\hat{\mathcal {E}}\), defined by (2.6), and it is comparable to \(\mathcal {E}\) with \(b=\alpha \), which was given in (6.1).
Proposition 6.6
For any \(u\in H_0^{\alpha /2}({\mathbb {R}}_*)\), we have
In addition
Proof
For notational simplicity, we start by denoting
so that, in particular, \(\mathcal {E}^o[u] = \lim _{t\rightarrow 0^+}\mathcal {E}^o_t[u]\). Using the symmetry of the semigroup \(P^o_t\), we write
We stress here that even though the constant function 1 is not in \(L^2({\mathbb {R}})\), it is still possible to use the symmetry of the operator \(P_t^o\), arguing by approximation. Therefore,
For any \(u\in C_c^\infty ({\mathbb {R}}_*)\), we have
Indeed, as \(u\in C_c^\infty ({\mathbb {R}}_*)\) we may assume that its support is contained in the interval \([\rho ,\rho ^{-1}]\) for some \(\rho \in (0,1)\). As \(\mathfrak {t}_{y,\mathfrak {f}}\ge \mathfrak {t}_{y,1}\) and \(\zeta ^o(t,y)=\zeta (t,y)=y+\zeta (t)\) for \(t<\mathfrak {t}_{y,1}\), we have
From elementary properties of a stable process we conclude that the right hand side vanishes as \(t\rightarrow 0+\). The above implies that for any \(u\in C_c^\infty ({\mathbb {R}}_*)\)
The second equality in (6.13) follows from the formula for the Dirichlet form of a symmetric stable process. \(\square \)
Fix y in \({\mathbb {R}}_*\) and let \(\{\tilde{\zeta }(t)\}_{t\ge 0}\) be the symmetric \(\alpha \)-stable Lévy process starting at y but killed at hitting 0. Clearly, its transition semigroup \(\{\tilde{P}_t\}_{t\ge 0}\), given as in (5.13), is made of symmetric Markov contractions on \(L^2({\mathbb {R}})\). In particular, its corresponding Dirichlet form equals:
for any function u belonging to its natural domain \(\mathcal {D}(\tilde{\mathcal {E}})=H_0^{\alpha /2}({\mathbb {R}}_*)\). For a proof see, e.g., [13, Section 3.3.3]. We consider as well \(\tilde{\mathcal E}_t[u]\) defined as in (6.11) with respect to \(\tilde{P}_t\). The same calculations that lead to (6.12) can be performed again to show that
Recalling the construction of the process \(\zeta ^o(t,y)\) in (3.19), we now write that
We then denote by \({\mathcal P}_m\) the family of sets \(\{\varepsilon _1,\ldots ,\varepsilon _{m}\}\subseteq \{-1,1\}^m\) such that \(\prod _{j=1}^m\varepsilon _j=1\) and by \({\mathcal P}_m^c\) its complement. Let
By a direct calculation,
It is then possible to show the following:
Lemma 6.7
The sequence \(\{s_m\}_{m\in {\mathbb {N}}}\) tends to 1/2. Moreover, if \(p_+>1/2\), then \(\{s_m\}_{m\in {\mathbb {N}}}\) is strictly decreasing, if \(p_+=1/2\), then it holds that \(s_m=1/2\) for any \(m\ge 1\) and if \(p_+<1/2\), then \(\{s_{2m-1}\}_{m\in {\mathbb {N}}}\) increases while \(\{s_{2m}\}_{m\in {\mathbb {N}}}\) decreases.
From (6.14), we now have that
If we suppose now that u belongs to \({\mathcal D}(\mathcal {E}^o)\), the closure of \(C^\infty _c({\mathbb {R}}_*)\) with respect to the form \(\mathcal {E}^o\), then (6.15) implies that u belongs to \(\mathcal {D}(\tilde{\mathcal {E}})=H_0^{\alpha /2}({\mathbb {R}}_*)\), as well. On the other hand, if u is in \(C_c^\infty ({\mathbb {R}}_*)\) then we can use the Hardy inequality (6.3) in (6.13) to show that \(\mathcal {E}^o[u] \preceq \tilde{\mathcal {E}}[u]\). Such an estimate then extends to \(H_0^{\alpha /2}({\mathbb {R}}_*)\) and implies in particular that \(H_0^{\alpha /2}({\mathbb {R}}_*)\subseteq \mathcal {D}(\mathcal {E}^o)\). We have thus proven that \(H_0^{\alpha /2}({\mathbb {R}}_*)= \mathcal {D}(\mathcal {E}^o)\) and there exists a positive constant C such that
which ends the proof of the proposition. \(\square \)
6.3 Proof of Theorem 6.5
In the present section, we are going to show that the Dirichlet forms \(\{\hat{\mathcal E}_{\lambda }\}_{\lambda >0}\), defined in (6.7), are M-convergent, when \(\lambda \) tends to \(+\infty \), to the Dirichlet form \(\bar{r}_*\hat{\mathcal E}\) given in (2.6)–(5.8). Thanks to Proposition 6.6, Theorem 6.5 will then follow immediately.
Recalling the meaning of the M-convergence in Definition 6.1, condition ii) easily follows choosing the trivial family \(\{v_\lambda \}_{\lambda > 0}\) of functions given by \(v_\lambda =u\) and using the Lebesgue dominated convergence theorem, together with (1.9) and (5.9). We now focus on showing condition i). Let \(\{u_{\lambda }\}_{\lambda >0}\) be a family of functions weakly convergent to u in \(L^2({\mathbb {R}})\). We start by noticing that if \(\liminf _{\lambda \rightarrow +\infty }\hat{\mathcal E}_{\lambda }[u_{\lambda }]=+\infty \), then condition i) clearly holds. We can then suppose without loss of generality that
Let \(\{\lambda _n\}_{n\in {\mathbb {N}}}\) be a sequence in \((0,+\infty )\) such that \(\lambda _n\rightarrow +\infty \) and
where we denoted \(u_n:=u_{\lambda _n}\), \(\hat{\mathcal E}_{n}:=\hat{\mathcal E}_{\lambda _n}\). We will use the following lemma, whose proof is presented in Sect. 6.4.
Lemma 6.8
Let \(\{f_n\}_{n\ge 1}\) be a bounded sequence in \(L^2({\mathbb {R}})\) such that \(\lim _{n\rightarrow \infty } \hat{\mathcal {E}}_n[f_n]<\infty \). Then, there exists a subsequence \(\{f_{n_k}\}_{k\ge 1}\) that is a.s. convergent to f.
We claim that u is in \(H_0^{\alpha /2}({\mathbb {R}}_*)\). Indeed, by Lemma 6.8 above, there exists a sub-sequence of \(\{u_n\}_{n\ge 1}\), that for simplicity we denote again by the same symbol, that is a.s. convergent in \({\mathbb {R}}\). From (5.10) and (1.9), we now notice that
Fatou’s lemma and (5.9) then imply that
From (6.16), we now have that \(u\in H^{\alpha /2}({\mathbb {R}})\). Using again (6.16), we notice that
for some \(N\in {\mathbb {N}}\). Assuming that \(\lambda ^{1/\alpha }_n|y|\ge 1\), we can use (5.2)–(5.7) to show that
The second case, where \(\lambda ^{1/\alpha }_n|y|< 1\), is trivial, since then
Thus, \(k_{\lambda _n}(y)\) can also be controlled from below by the term appearing on the utmost right hand side of (6.18). From (6.17) and (6.18), it follows that
for any \(n\ge N\). We shall use the following.
Lemma 6.9
Let \(\{f_n\}_{n\ge 1}\) be a bounded sequence in \(L^2({\mathbb {R}})\) such that \(\lim _{n\rightarrow \infty } \hat{\mathcal {E}}_n[f_n]<\infty \) and the inequality (6.19) holds. Then, for any \(\rho \) in \((0,1-1/\alpha )\),
The lemma is shown in Sect. 6.5. We proceed with its application to the proof of Theorem 6.5. Using Fatou’s lemma in (6.20), we conclude that
for any \(\rho \) in \((0,1-1/\alpha )\). Thanks to Proposition 6.3 and Corollary 6.4, we then infer that u is in \(H^{ \alpha /2}_0({\mathbb {R}}_*)\). Since \(\{u_n\}_{n\ge 1}\) is a.s. convergent to a function u in \(H^{ \alpha /2}_0({\mathbb {R}}_*)\), we can use Fatou’s lemma and (5.9) to write that:
and we have proven part i) of Definition 6.1, which ends the proof of Theorem 6.5.
What remains to be done is to prove the auxiliary results presented above. Before doing so, we however need the following result:
Lemma 6.10
Let \(\{f_n\}_{n\ge 1}\) be bounded in \(L^2({\mathbb {R}})\) such that \(\lim _{n\rightarrow \infty } \hat{\mathcal {E}}_n[f_n]<\infty \). Then,
uniformly in n. Here we have denoted by \(\hat{f}(\cdot )\) the Fourier transform of a function \(f\in L^2({\mathbb {R}})\).
Proof
Thanks to (1.9), we know that \(p_*:= \inf _{k\in {\mathbb {T}}} p_+(k) \, > \, 0\). Since \(\lim _{n\rightarrow \infty } \hat{\mathcal {E}}_n[f_n]<\infty \), we then conclude from (6.7) that
where \(\bar{r}_{\lambda }\) is given by (5.6). The right-hand side is then of the same order of magnitude as
Fixing \(K>0\), it now follows that
where \(\Psi _n:=\Psi _{\lambda _n}\) was defined in (5.24). Thus, for any n so large that \( K\le \lambda _n^{1/\alpha }\), the estimates (6.21), (5.25) and (5.27) imply that
where \(\theta _*\) is given by (5.26). Hence,
and the conclusion of Lemma 6.10 follows. \(\square \)
6.4 Proof of Lemma 6.8
Fix \(\delta >0\) and let us consider the function \(\phi _\delta \) given in (5.36). We claim that the sequence \(\{f_n\phi _\delta \}_{n\ge 1}\) is strongly compact in \(L^2({\mathbb {R}})\). Using the Pego Criterion, see [29, Theorem 3], it is enough to show that
uniformly in \(n \in {\mathbb {N}}\). Clearly, (6.22) holds since the sequence is uniformly bounded in \(L^2({\mathbb {R}})\). To show (6.23), we notice that
where for any fixed \(L>0\), we have
The Cauchy-Schwartz inequality and Lemma 6.10 then imply that for any fixed \(L>0\),
On the other hand, recalling that \(\{f_n\}_{n\ge 1}\) is uniformly bounded in \(L^2({\mathbb {R}})\) and \(\hat{\phi }_\delta \) belongs to the Schwartz class, we conclude that for any \(\varepsilon >0\) it is possible to choose \(L:=L(\varepsilon )\), sufficiently large, such that
uniformly in n. From (6.24)–(6.26), we can then conclude that (6.23) follows. From the compactness of \(\left\{ f_n\phi _\delta \right\} _{n\ge 1}\), we can now choose a subsequence that is a.e. convergent on \([-\delta ,\delta ]\). Using the Cantor diagonal argument, we then find a subsequence of \(\left\{ f_n\right\} _{n\ge 1}\) that converges a.e. on \({\mathbb {R}}\). \(\square \)
6.5 Proof of Lemma 6.9
First observe that since the sequence \(\{f_n\}_{n\ge 1}\) is bounded in \(L^2({\mathbb {R}})\), it is enough to show that for any \(\rho \) small enough, we have:
We will actually prove an analogue of (6.27) with the integral over [0, 1], as the argument in the case of \([-1,0]\) is similar. The proof is divided into three steps.
Step 1. We start by claiming that for any \(\rho \) in (0, 1), \(\gamma >1\) and \(\lambda _n\) large enough,
where
and \(\log _2x:=\log \log x\), \(x>1\). Indeed, let us denote
for any \(m\in {\mathbb {N}}\) and \(I_{0}:=\left\{ y\in [0,1] :y<\lambda _n^{-1/\alpha }\right\} \). Estimate (6.19) now implies that
where a positive integer \(N_1 \) is such that
and
It is not difficult to check now that the sequence \(\{ a_m\}_{m\ge 1}\) is decreasing and thus, its minimum in \(1\le m\le m_*(\gamma ):=N_1-[\log ^{\gamma }N_1]\) equals \(a_{m_*(\gamma )}\). Hence, (6.30) implies that
since \( e^{m_*(\gamma )}\lambda _n^{-1/\alpha } \ge e^{m_*(\gamma )-N_1}\) (see (6.31)). Estimate (6.28) follows from an observation that \(e^{m_*(\gamma )-N_1} \ge c_n(\gamma )\) and
In order to prove the utmost right inequality, it is enough to show that
To see the last estimate, recall that by (6.31), we have \(N_1-1\le \log \lambda _n^{1/\alpha }\le N_1.\) Hence it suffices to show \(\rho \alpha [\log ^\gamma N_1]-\kappa \log N_1 \, \succeq \, (\log N_1)^{\gamma /2}\). The latter holds, since \(\gamma >1\vee \gamma /2\).
Step 2. We then show that for any \(\lambda _n>1\), the function \(f_n\) can be decomposed as:
for two functions \(f^{(1)}_n\), \(f^{(2)}_n\) such that
We start by defining
Clearly, (6.32) then holds. We now use again the function \(\Psi _n\) defined in (5.24). Fixed \(h>0\), we apply (5.27) to write that
Using the Cauchy-Schwartz inequality and then (6.21), we conclude that estimate (6.33) holds:
On the other hand, estimate (6.34) follows from (5.25) and (6.21):
Step 3. Fix n large enough so that \(\lambda _n\ge e\) and \(\gamma >1\), we consider
for any \(m\in {\mathbb {N}}_0\). We can then write
where \(N_2\) is the (unique) positive integer such that \(N_2 c_n(\gamma )\le 1< (N_2+1)c_n(\gamma )\) and
To control the first term \(I_n\), we start by noticing that if \(|y|\le c_n(\gamma )\), then \( \lambda _n^{-1/\alpha }+y \le 2c_n (\gamma )\) for sufficiently large n. We can then write:
provided that \(\rho \) is so small that \( \alpha (1-\rho )>1\). From (6.28) we thus have that \(I_n\preceq c_n(\gamma /2)\preceq 1\), uniformly in n. Concerning the second term \(I\!I_n\), we use (6.32) to decompose it as \(I\!I^{(1)}_n+I\!I^{(2)}_n\) where for any \(j=1,2\), \(I\!I^{(j)}_n\) is obtained from \(I\!I_n\) by replacing there \(f_n\) with \(f^{(j)}_n\). Noticing that \( N_2= \left[ c^{-1}_n(\gamma )\right] \), Estimate (6.33) now implies that
To control \(I\!I^{(2)}_n\), we recall that \(\lambda _n \) dominates \(c_n^{-\alpha (1-\rho )}(\gamma ) = \exp \left\{ \alpha (1-\rho )\log _2^{\gamma }\lambda _n \right\} \) (cf. (6.29)) for n sufficiently large, and then, thanks to (6.34), we write
This ends the proof of (6.27). \(\square \)
Notes
Assumptions made in the present paper hold, e.g., when the linear kinetic equation is obtained as the limit of the Wigner functions corresponding to a harmonic chain with momenta randomly exchanged at Poissonian times. See, e.g., [21, Section 2.1.3] The derivation of the corresponding kinetic equation follows the same argument as in [6].
References
Aceves-Sánchez, P., Schmeiser, C.: Fractional diffusion limit of a linear kinetic equation in bounded domain. Kinetic Related Models 10, 541–551 (2017)
Applebaum, D.: Lévy Processes and Stochastic Calculus. Cambridge University Press, Cambridge (2004)
Bardos, C., Golse, F., Moyano, I.: Linear Boltzmann equation and fractional diffusion. KRM 11, 1011–1036 (2018)
Basile, G., Bovier, A.: Convergence of a kinetic equation to a fractional diffusion equation. Markov Process. Related Fields 16, 15–44 (2010)
Basile, G., Komorowski, T., Olla, S.: Diffusion limit for a kinetic equation with a thermostatted interface. Kinetic Related Models 12, 1185–1196 (2019). https://doi.org/10.3934/krm.2019045
Basile, G., Olla, S., Spohn, H.: Energy transport in stochastically perturbed lattice dynamics. Arch. Ration. Mech. 195(1), 171–203 (2009)
Billingsley, P.: Convergence of Probability Measures. Wiley Series in Probability and Statistics: Probability and Statistics, 2nd edn. Wiley, New York (1999)
Bobrowski, A., Komorowski, T.: Diffusion approximation for a simple kinetic model with asymmetric interface. J. Evol. Equ. 22, 42 (2022). https://doi.org/10.1007/s00028-022-00801-x
Cesbron, L.: Anomalous diffusion limit of kinetic equations in spatially bounded domains. Commun. Math. Phys. 364, 233–286 (2018)
Cesbron, L.: Fractional diffusion limit of a linear Boltzmann model with reflective boundaries in a half-space, preprint arXiv (2020)
Cesbron, L., Mellet, A., Puel, M.: Fractional diffusion limit for a kinetic equation in the upper-half space with diffusive boundary conditions. Arch. Ration. Mech. Anal. 235, 1245–1288 (2020)
Cesbron, L., Mellet, A., Puel, M.: Fractional Diffusion limit of a kinetic equation with diffusive boundary conditions in a bounded interval. arXiv:2107.01011v1
Chen, Z.-Q., Fukushima, M.: Symmetric Markov Processes, Time Change, and Boundary Theory, London Mathematical Society Monographs Series, 35. Princeton University Press, Princeton (2012)
Dyda, B.: A fractional order Hardy inequality. Ill. J. Math. 48, 575–588 (2004)
Ethier, S.N., Kurtz, T.G.: Markov Processes. Characterization and Convergence. Wiley Ser. Probab. Math. Statist., New York (1986)
Fukushima, M., Oshima, Y., Takeda, M.: Dirichlet forms and symmetric Markov processes. De Gruyter Studies in Mathematics, vol. 19. Berlin, New York (2011)
Gnedenko, B.V., Kolmogorov, A.N.: Limit distributions for sums of independent random variables. Translated and annotated by K. L. Chung. With an Appendix by J. L. Doob. Addison-Wesley Publishing Company, Inc., Cambridge (1954)
Iksanov, A.; Pilipenko, A.: On a skew stable Lévy process, 23 pp. arXiv (2021)
Jara, M., Komorowski, T.: Limit theorems for some continuous-time random walks. Adv. Appl. Probab. 43, 782–813 (2011)
Jara, M., Komorowski, T., Olla, S.: Limit theorems for additive functionals of a Markov chain. Ann. Appl. Prob. 19, 2270–2300 (2009)
Jara, M., Komorowski, T., Olla, S.: Superdiffusion of energy in a chain of harmonic oscillators with noise. Commun. Math. Phys. 339, 407–453 (2015)
Komorowski, T., Olla, S., Ryzhik, L.: Fractional diffusion limit for a kinetic equation with an interface. Ann. Prob. 48, 2290–2322 (2020)
Komorowski, T., Olla, S., Ryzhik, L., Spohn, H.: High frequency limit for a chain of harmonic oscillators with a point Langevin thermostat. Arch. Ration. Mech. Anal. 237, 497–543 (2020)
Komorowski, T., Olla, S.: Kinetic limit for a chain of harmonic oscillators with a point Langevin thermostat. J. Funct. Anal. 279, Article # 108764 (2020). https://doi.org/10.1016/j.jfa.2020.108764
Komorowski, T., Olla, S.: Thermal boundaries in kinetic and hydrodynamic limits. In: Salvarani, F. (ed.) Recent Advances in Kinetic Equations and Applications, Springer INdAM, vol. 48, Series, pp. 253–288 (2021). ISBN-13: 978-3030829452
Mellet, A.: Anomalous diffusion phenomena: a kinetic approach, Séminaire Laurent Schwartz–Équations aux dérivées partielles et applications. Année 2014–2015, Exp. No. XII, 16 pp., Ed. c. Polytech., Palaiseau (2016)
Mosco, U.: Composite media and asymptotic Dirichlet forms. J. Funct. Anal. 123, 368–421 (1994)
Muramatu, T.: On imbedding theorems for Sobolev spaces and some of their generalizations. Publ. Res. Inst. Math. Sci. Ser. A 3, 393–416 (1967/1968)
Pego, R.L.: Compactness in \(L^2\) and the Fourier Transform. Proc. Am. Math. Soc. 95(2), 252–254 (1985)
Sato, K.: Lévy Processes and Infinitely Divisible Distributions. Cambridge Studies in Advanced Mathematics, vol. 68. Cambridge University Press, Cambridge (1999)
Whitt, W.: Stochastic Process Limits: An Introduction to Stochastic-Process Limits and their Application to Queues. Springer, New York (2002)
Whitt, W.: Internet Supplement to Stochastic-Process Limits. Available at www.columbia.edu/ ww2040/supplementno.pdf
Acknowledgements
The authors would like to thank the two referees for their careful reading and comments. They lead to a significant improvement of the manuscript.
Author information
Authors and Affiliations
Contributions
TK proposed the model and most techniques. KB proposed the use of Dirichlet forms. LM critically corrected the development. All authors reviewed and edited the manuscript.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
K.B. was supported through the DFG-NCN Beethoven Classic 3 programme, Contract No. 2018/31/G/ST1/02252 (National Science Center, Poland) and SCHI-419/11-1 (DFG, Germany). T.K. acknowledges the support of the NCN Grant 2020/37/B/ST1/00426.
Appendices
Appendix A. Proofs of the results of Sect. 6.1
1.1 A.1. Proof of Proposition 6.2
If we assume that \(u\in C_c^\infty ({\mathbb {R}}_*)\), then inequality (6.3) follows from [14, Theorem 1.1 part T4)]. By Fatou’s lemma, we can then extend (6.3) to all functions u in \(H_0^{b/2}({\mathbb {R}}_*)\) with the same constant. The constant \(C_\alpha \) is bounded on compact subsets of (1, 2), which follows from the inspection of the constant in the last inequality of the proof of [14, Lemma 3.3], see also the proof of [14, Theorem 1.1]. \(\square \)
1.2 A.2. Proof of Proposition 6.3
For simplicity, let us denote the set on the right hand side of (6.4) by \({\mathcal H}\). We start by noting that Hardy inequality (6.3) immediately implies that \( H_0^{b/2}({\mathbb {R}}_*)\subset {\mathcal H}\). In order to show the other inclusion, let us take \(u\in \mathcal {H}\). For any \(\delta >0\), let us consider again the function \(\phi _\delta \), as considered in (5.36), and define \(\psi _\delta =1-\phi _\delta \). Without any loss of generality, we can also assume that \(\Vert \psi _\delta '\Vert _{\infty }\le 2/\delta \). Let \(u_\delta :=u\psi _\delta \). It is easy to check that \(u_\delta \) is in \(\mathcal {H}\) and that it can be approximated by elements of \(C_c^\infty ({\mathbb {R}}_*)\) in \(H^{b/2}({\mathbb {R}})\)-norm. To conclude, it is then enough to show that \(\lim _{\delta \rightarrow 0}\Vert v_\delta \Vert _{H^{b/2}({\mathbb {R}})} = 0\) where we denoted \(v_\delta :=u_\delta -u\). Clearly, \(\Vert v_\delta \Vert _{L^2}\rightarrow 0\) as \(\delta \) goes to 0, by the dominated convergence theorem. On the other hand, \(\mathcal {E}[v_\delta ]\) can be decomposed as
To control the first term \(I_\delta \), we notice that for any fixed \(\varepsilon >0\), there exists \(\delta :=\delta (\epsilon )>0\) sufficiently small such that
To deal with the second term \(I\!I_\delta \), we start by splitting it as
Recalling that \(v_\delta (y) = 0\) if \(|y|\ge 2\delta \), we denote \(A=\{yy'>0, |y-y'|\ge 4\delta , |y|\le 2\delta \}\) so that
provided that \(\delta :=\delta (\epsilon )>0\) is sufficiently small. Similarly, we also have that
where we have denoted
Recalling that \(u\in \mathcal {H}\), it follows that for any sufficiently small \(\delta >0\),
Since we have chosen \(\psi _\delta \) such that \(\Vert \psi _\delta '\Vert _{\infty }\le 2/\delta \), we have that
for sufficiently small \(\delta >0\). Summarising, estimates (A.1) – (A.4) imply that \(\mathcal {E}[v_\delta ]<\epsilon \) for any \(\delta \) small enough, and the conclusion of the proof then follows. \(\square \)
1.3 A.3. Proof of Corollary 6.4
First, we show (6.5). For simplicity, let us denote the set appearing on its right-hand side by \(\mathcal {H}'\). Then, the inclusion \(H_0^{b_0/2}({\mathbb {R}}_*)\subset \mathcal {H}'\) trivially follows once we show that
for any \(b<b_0\) in (0, 2) and u in \(H^{b_0/2}({\mathbb {R}})\). If \(u\in C^\infty _c({\mathbb {R}})\), then the above inequality immediately follows from
The general conclusion can be then obtained by an approximation argument. To show the other inclusion, let us take now a function u in \(\mathcal {H}'\). Proposition 6.2 and estimate (A.5) then imply that for any \((1+b_0)/2\le b\le b_0\), we have
Thanks to Fatou’s lemma and Proposition 6.3, it follows that \(u\in H_0^{b_0/2}({\mathbb {R}}_*)\). We have thus showed (6.5). To prove (6.6), let us suppose that \(u\in H^{b_0/2}({\mathbb {R}})\) satisfies \(u(0)=0\). From (6.5) and Proposition 6.3, it suffices to show that
for any \(b\in (1,b_0)\). Since \(b_0>1\), we recall from [28, Theorem 1, p. 394] that u possesses an Hölder continuous representative and that in particular,
It then follows that
and the proof of the corollary is concluded. \(\square \)
Appendix B. Proof of Proposition 3.7
1.1 B.1. Representation of a weak solution
We will assume without loss of generality that \(T_\infty =0\). We also suppose for the moment that \(T_o=0\). Then Proposition 3.7 can be obtained form the observation that
is the semigroup solution to Cauchy problem (1.17), where \(P^o_t\) and \(\bar{\theta }\) have been defined in (5.13) and (3.11), respectively.
If \(T_o\ne 0\), let us consider again the smooth function \(\phi _M\) given in (5.36). Since \(\bar{W}_0-T_o\phi _M\) is in \(L^2({\mathbb {R}})\cap \mathcal {H}_o\) and \(\big (\bar{W}_0-T_o\phi _M\big )(0)=0\), we already know that
is a weak solution of (1.17) for \(T_o=0\). Thus, \(\bar{W}_M\in C\left( [0,+\infty );L^2({\mathbb {R}})\right) \cap L^2_\textrm{loc}\left( [0,+\infty ); \mathcal {H}_o\right) \) and
It is not difficult to check now that \(\bar{W}_0-T_o\phi _M\rightarrow \bar{W}_0-T_o\), as \(M\rightarrow +\infty \), both in \(\mathcal {H}_o\) and pointwise. Recalling that \(P^o_t\) is the Markov semigroup associated with the form \(\hat{\mathcal {E}}\), we have that
It then follows that \(\bar{W}_M\rightarrow \bar{W}-T_o\) in \(\mathcal {H}_o\) and \(\bar{W}-T_o\) is in \(L^2_{\text {loc}}(0,+\infty ,\mathcal {H}_o)\). Finally, we can follow the same arguments in the proof of Lemma 5.6 to show that
for any F in \(C^\infty _c({\mathbb {R}})\). Hence, we can pass to the limit in (B.1) and we conclude that \(\bar{W}(t,y)\) given by (3.22) is a weak solution in the sense of Definition 2.4.
1.2 B.2. Uniqueness of a weak solution
For notational simplicity, let us denote by \(\langle \cdot ,\cdot \rangle _{\mathcal {H}_o}\) the scalar product on \(\mathcal {H}_o\). Let \(\bar{W}_1\), \(\bar{W}_2\) be two weak solutions (in the sense of Definition 2.4) of Cauchy problem (1.17). By definition, we know that for each \(j=1,2\), there exists \(T^j_\infty \) such that \(\bar{W}_j-T^j_\infty \in C([0,+\infty );L^2({\mathbb {R}}))\) and \(\bar{W}_1(0)=\bar{W}_2(0)=\bar{W}_0\). It then follows immediately that \(T^1_\infty =T^2_\infty \). This in turn implies that
Assuming for the moment that \(\bar{W}\) belongs to \(C^1\left( [0,+\infty ); \mathcal {H}_o\right) \), it is easy to check that
In the more general case where \(\bar{W}\) is in \(L^2_\textrm{loc}\left( [0,+\infty ); \mathcal {H}_o \right) \), (B.2) follows by a mollification argument in time. Recalling that \(\bar{W}\) satisfies (2.7) for \(\bar{W}_0=T_o=0\), we can now deduce that for any \(F \in L^2({\mathbb {R}})\cap \mathcal {H}_o\), it holds that
In particular, from (B.2) and (B.3) with \(F=\bar{W}(s)\), we finally get that
which implies that \(\bar{W} (t)=0\). \(\square \)
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Bogdan, K., Komorowski, T. & Marino, L. Anomalous diffusion limit for a kinetic equation with a thermostatted interface. Probab. Theory Relat. Fields 189, 721–769 (2024). https://doi.org/10.1007/s00440-023-01251-3
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00440-023-01251-3
Keywords
- Fractional diffusion limit from kinetic equation
- Fractional Laplacian with boundary condition
- Stable processes with interface