Abstract
In this work, we consider a heat equation with a nonlinear term of polynomial type and with two different cases in the diffusion term. The first case (anomalous diffusion) concerns to the fractional Laplacian operator with parameter \(1<\alpha <2\), while the second case (classical diffusion) involves the classical Laplacian operator. When \(\alpha \rightarrow 2^{-}\), we prove the uniform convergence of strong solutions of the anomalous diffusion case to a strong solution of the classical diffusion case. Moreover, we rigorously derive a convergence rate (when \(\alpha \rightarrow 2^{-}\)) which highlights some phenomenological effects closely related to the structure of solutions.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
In this paper, we consider the following multi-dimensional, nonlinear and anomalous diffusion heat equation in the whole space \(\mathbb {R}^n\) with \(n \ge 1\):
Here, the function \(u:[0,+\infty [\times \mathbb {R}^n\rightarrow \mathbb {R}\) is the solution, and \((-\Delta )^{\alpha /2}\, u\) is the anomalous diffusion term which is given by the fractional Laplacian operator \((-\Delta )^{\alpha /2}\). We recall that this operator is defined in the Fourier level by the expression
Moreover, in the spatial variable, the fractional Laplacian operator is defined as the following non-local operator:
where \({\textbf {p.v.}}\) denotes the principal value and \(c_{n,\alpha }>0\) is a constant depending on the dimension n and the parameter \(\alpha \). Finally, \(\eta \in \mathbb {R}^n\) is a fixed vector, and moreover, the parameter \(b \in \mathbb {N}\) in the nonlinear term verifies \(b\ge 2\).
We may observe that this highly nonlinear term essentially behaves as the derivative of a polynomial of degree b in the variable u. Thus, this term agrees with the classical assumption for the non-linearity in the qualitative study of the heat equation. See, for instance, [3,4,5,6, 12] and the references therein.
Nonlinear evolution PDEs involving the fractional Laplacian, which describe the anomalous or \(\alpha -\)Lévy stable diffusion, have been extensively studied in the physical and mathematical points of view. From the physical point of view, and for \(b=2\), the equation (1) deals with a generalized Burgers-type equation [4] which has been largely used to model a variety of physical phenomena such as, for example, the anomalous homogeneous turbulence [10], applications to hydrodynamics and statistical mechanics [17], and moreover, applications to molecular biology in the modeling of growth of molecular interfaces [20]. In the latter application, the general algebraic non-linear term \(u^b\), with \(b \in \mathbb {N}^*\) and \(b\ge 2\), provides a good model for multi-particle interactions. For more references, see the book [16].
From the mathematical point of view, when the solution \(u(t,\cdot )\) is considered as the density of a probability distribution for every \(t > 0\), the equation (1) has an important probabilistic interpretation in the theory of nonlinear Markov processes and propagation of chaos. See, e.g., the works [11, 14] and the references therein.
Getting back to the expression (1), we observe that for each value of the parameter \(1<\alpha <2\) in the fractional Laplacian operator \((-\Delta )^{\alpha /2}\) we get a corresponding fractional PDE. We thus denote by \(u_\alpha (t,x)\) the corresponding solution of each equation and the main objective of this paper is to understand the asymptotic behavior of the family of functions \(u_\alpha (t,x)\) when the parameter \(\alpha \) goes to 2. This question was pointed out from the experimental point of view in [10, 20] and has some interesting applications in these physical and biological models. Our aim is then to provide a rigorous mathematical framework to give an answer.
Formally, we may observe that if in the expression (1) we set \(\alpha =2\), then we get a classical diffusion equation involving the Laplacian operator:
Consequently, if \(u_2(t,x)\) denotes a solution of the equation above, we are interested in providing a rigorous understanding of the expected convergence \(u_\alpha (t,x) \rightarrow u_2(t,x)\), when \(\alpha \rightarrow 2\). It is worth mentioning although this problem is easily formulated, it is not a trivial study since for each value of the parameter \(\alpha \) we have a different fractional PDE depending on this parameter.
In the particular case of the following linear equation in a smooth and bounded domain \(\Omega \subset \mathbb {R}^n\):
and where the function \(f_\alpha (t,x)\) does not depend on the solution \(u_\alpha \), this convergence problem was studied by U. Biccari & V. Hernández-Santamaría in [2]. For a time \(0<T<+\infty \), the authors consider a family of functions \(f_\alpha \in L^2(0,T, H^{-\alpha }(\Omega ))\), which is uniformly bounded with respect to the parameter \(\alpha \): \(\Vert f_\alpha (t,\cdot )\Vert _{H^{-\alpha } (\Omega )} \le C \), and such that when \(\alpha \rightarrow 2\) we have the convergence \(f_\alpha (t,\cdot )\rightarrow f(t,\cdot )\) in the weak topology of the space \(H^{-1}(\Omega )\). Then, by using a compactness argument (due to the boundness of the domain \(\Omega \)) it is shown that weak solutions of equation (3) converge in the strong topology of the space \(L^2(0,T, H^{1-\delta }_{0}(\Omega ))\) (with \(0<\delta \le 1\)) to a weak solution of the corresponding linear heat equation with datum f.
On the other hand, L. Ignat & J.D. Rossi studied in [12], among other things, that weak solutions u(t, x) fo the nonlinear heat equation (2) can be obtained as the limit (when \(\varepsilon \rightarrow 0^{+}\)) of the weak solutions to the following nonlocal convection-diffusion equation in the whole space \(\mathbb {R}^n\):
This equation has the same scaling properties of the equation (2) and here, for suitable non-negative functions \(J\in \mathcal {S}(\mathbb {R}^n)\) and \(G \in \mathcal {S}(\mathbb {R}^n)\), we have \(J_\varepsilon (x)= \frac{1}{\varepsilon ^n} J (x / \varepsilon ) \) and \(G_\varepsilon (x)= \frac{1}{\varepsilon ^n} G (x / \varepsilon )\) respectively. Moreover, J is a radially symmetric function and the key assumption is that its Fourier transform \(\widehat{J}(\xi )\) satisfies the following condition:
which is similarly satisfied for the symbol \(\vert \xi \vert ^2\) of the classical Laplacian operator. In this setting, by using some sharp estimates of the kernel associated to the linear problem, and moreover, by setting the vector \(\eta =(\eta _1, \ldots , \eta _n)\) in the equation (2) as \(\displaystyle {\eta _i= \int _{\mathbb {R}^n} x_i G(x)dx}\), for all time \(0<T<+\infty \) it is proven the following convergence result in the natural framework (due to the Plancherel’s identity) of the Lebesgue space \(L^2(\mathbb {R}^n)\):
Nevertheless, we remark that this result cannot be applied to the case of the equation (1). Indeed, since the symbol \(\vert \xi \vert ^\alpha \) of the operator \((-\Delta )^{\alpha / 2}\) does not verify the key condition (5) the nonlocal diffusion operator \(1/\varepsilon ^2(J_\varepsilon *(\cdot ) - I_d)\) (where \(I_d\) is the identity operator) does not contain the fractional Laplacian operator \((-\Delta )^{\alpha / 2}\) as a particular case. Moreover, one the main property of the approximated equation (4) is the same scaling of the equation (2), which is not the case of the equation (1) due to the different scaling provided by the fractional Laplacian operator.
In this work, we will use a different approach. For any time \(0<T<+\infty \), in the framework the space \(L^{\infty }([0,T]\times \mathbb {R}^n)\) we shall study the convergence (in the strong topology) of the strong (mild) solutions \(u_{\alpha }(t,x)\) for the anomalous diffusion equation (1) (given in the expression (14)) to a strong solution \(u_2(t,x)\) for the classical diffusion equation (2) (given in the expression (15)). See our main result given in Theorem 2.2 for the details. This uniform convergence also allows us to prove a strong convergence in the \(L^{p}_{t} L^{q}_{x}\) spaces (see the Corollary 2.1).
Our method is based on two key ideas. On the one hand, we study the convergence of the fundamental solution \(p_\alpha (t,x) \) associated with the fractional linear heat equation (see the expression (16) for a definition) to the heat kernel h(t, x). On the other hand, we prove some uniform estimates with respect to the parameter \(\alpha \) for the family of functions \(u_{\alpha }(t,x)\).
Finally, we think that in a further research our method could be adapted to the case when the fractional Laplacian operator \((-\Delta )^{\alpha /2}\) in the equation (1) is substituted by a more general Lévy-type operator \(\mathcal {L}^\alpha \). For a definition and some well-known properties of this latter operator, we refer to the book [13].
2 The main result
Let us consider the Cauchy problem for both anomalous (when \(1<\alpha <2\)) and classical (when \(\alpha =2\)) nonlinear heat equation:
Well-posedness (WP) issues for this equation have been studied in several works [3, 8, 9] and it is well-known that for an initial datum \(u_{0} \in L^1(\mathbb {R}^n)\) the initial value problem (7) has a unique solution \(u_\alpha \in \mathcal {C}([0,+\infty [, L^1(\mathbb {R}^n))\) which verifies
Moreover, for \(1\le p \le +\infty \) this solution also verifies \(\displaystyle {u_\alpha \in \mathcal {C}(]0,+\infty [, W^{1,p}(\mathbb {R}^n))}\), and the following estimate holds:
Finally, under the additional assumption on the initial datum: \(u_{0} \in L^1 \cap L^p(\mathbb {R}^n)\) the corresponding solution verifies \(\displaystyle {u_\alpha \in \mathcal {C}([0,+\infty [, L^p(\mathbb {R}^n))}\), and for all time \(t\ge 0\) we have the estimate
As mentioned, our aim is to study the convergence (when \(\alpha \rightarrow 2\)) of mild solutions of the equation (7). For technical reasons, principally due to the study of the limit concerning the highly nonlinear term \(\nabla (u^{b}_{\alpha }) \rightarrow \nabla (u^{b}_{2}) \) (recall that b is any integer such that \(b \ge 2\)), we shall need more regularity than the one given by the space \(W^{1,p}(\mathbb {R}^n)\). For this, we shall consider initial data belonging to the space \(L^1 (\mathbb {R}^n) \cap H^s(\mathbb {R}^n)\) with \(s > n / 2\).
The global-well posedness in the setting of the space \(L^1 (\mathbb {R}^n) \cap H^s(\mathbb {R}^n)\) is rather standard but, to our knowledge, this fact has not been proven before. Consequently, only for the completeness of this paper, we state and we will give a proof of the following theorem. We emphasize that the only novelty is the gain of regularity given in the expression (9) below, which follows from the additional hypothesis \(u_0 \in H^s(\mathbb {R}^n)\) with \(s>n/2\).
Theorem 2.1
Let \(1<\alpha \le 2\). For \(s>n/2\), let \(u_{0} \in L^1 \cap H^s(\mathbb {R}^n) \) be an initial datum. Then there exists a unique mild solution
of the equation (7). Moreover, this solution is regular:
and it verifies the equation (7) in the classical sense.
We study now the convergence of mild solutions for the equation (7) when \(\alpha \rightarrow 2^{-}\). For each \(1<\alpha <2\) we consider \(u_{0,\alpha } \in L^1\cap H^s(\mathbb {R}^n)\) an initial datum and we shall denote by \(u_\alpha (t,x)\) its corresponding arising solution of the equation (7), which is given by Theorem 2.1. Moreover, for the classical case (when \(\alpha =2\)) we similarly consider an initial datum and its corresponding solution \(u_{0,2} \in L^1\cap H^s(\mathbb {R}^n)\) and \(u_2(t,x)\) respectively.
We shall assume the following strong convergence on the initial data:
By the Sobolev embedding \(H^s(\mathbb {R}^n)\subset L^{\infty }(\mathbb {R}^n)\) (since \(s>n/2\)) we also have \(u_{\alpha ,0} \in L^{\infty }(\mathbb {R}^n)\), \(u_{0,2}\in L^{\infty }(\mathbb {R}^n)\) and the convergence above also holds true in the space \(L^{\infty }(\mathbb {R}^n)\). Thus, for the corresponding family of solutions \((u_\alpha )_{1<\alpha \le 2}\) we will study the uniform convergence for any \( 0<T<+\infty \):
Moreover, we are also interested in studying the convergence rate in (11). For this, we introduce a parameter \(\gamma >0\) and we shall assume the estimate (12) below, which is a given convergence rate of the initial data in the space \(L^\infty (\mathbb {R}^n)\). Our aim is then to study when the family of solutions follows this prescribed convergence rate. In this setting, our main result reads as follows:
Theorem 2.2
Let \((u_{0,\alpha })_{1<\alpha \le 2}\) be a family of initial data such that for all \(1<\alpha \le 2\) we have \(u_{0,\alpha } \in L^1 \cap H^s(\mathbb {R}^n)\). Let \( (u_\alpha )_{1<\alpha \le 2}\) be corresponding family of solutions to the equation (7) given by Theorem 2.1.
We assume the convergence given in (10), and moreover, for a parameter \(\gamma >0\) we assume the estimate
where \(\textbf{c}>0\) is a given generic constant. Then, there exists \(0<\varepsilon \ll 1\), and there exists a constant \(\textbf{C}>0\), which depends on the parameters \(\eta \) and b in the equation (7), the initial data \(u_{0,2}\), the quantity \(\varepsilon \) and the constant \(\textbf{c}\), such that the following estimate holds:
for all \(1+\varepsilon< \alpha < 2\) and for all \( 0<T<+\infty \).
Some remarks are in order here. First note that our approach allows us obtain a uniform convergence (in both the temporal and the spatial variables) which is not studied in the previous related works [2] and [12]. Moreover, it is interesting to observe that the convergence rate given in the estimate (13) is determined by a competition between the quantities \((2-\alpha )^\gamma \) and \((2-\alpha )\).
In order to make a deeper discussion of this fact, let us briefly explain the general idea of the proof. As pointed out, we shall consider mild solutions of the equation (7). Thus, for \(1<\alpha <2\) we have
where the kernel \(p_\alpha (t,x)\) is given in (16), while for \(\alpha =2\) we have
where h(t, x) denotes the well-known heat kernel. The estimate (13) is then obtained by the following estimates
and
on the linear and the nonlinear terms respectively. For the sake of simplicity, we will only explain more in detail the estimates on the linear term. Of course the estimates for the nonlinear term are much more delicate, but they follow some similar ideas. We split the linear term as
where we have
and
Here, the quantity \((2-\alpha )^\gamma \) is the convergence rate assumed for the initial data, while the quantity \((2-\alpha )\) is the convergence rate of the kernels \(p_\alpha (t,x) \rightarrow h(t,x)\) (when \(\alpha \rightarrow 2^{-}\)), which is rigorously proven in Lemma 5.1 below.
Since we have \(1< \alpha <2\) and therefore \(0<2-\alpha < 1\), the estimate (13) yields the following conclusions by considering two cases of the parameter \(\gamma \).
-
The case \(0<\gamma \le 1\). Here we have \(\displaystyle {\max \Big ((2-\alpha )^\gamma , 2-\alpha \Big ) = (2-\alpha )^\gamma }\), and consequently, the solutions \(u_\alpha (t,x)\) converge to the solution \(u_2(t,x)\) with the same convergence rate as that of the initial data.
-
The case \(\gamma > 1\). In this case we have \(\displaystyle {\max \Big ((2-\alpha )^\gamma , 2-\alpha \Big ) = 2-\alpha }\). Then, it is interesting to observe that the convergence rate of the solutions does not follow the one of initial data. More precisely, the solutions \(u_\alpha (t,x)\) converge to the solution \(u_2(t,x)\) with a rate of order \(2-\alpha \), which is slower than the convergence rate of the initial data \((2-\alpha )^\gamma \).
Summarizing, the increasing of the parameter parameter \(\gamma \) makes the assumption (12) strong but not the result given in (13). This is an interesting phenomenological effect, which is given by the convergence rate of the kernels \(p_{\alpha }(t,\cdot ) \rightarrow h(t,\cdot )\).
On the other hand, as mentioned in the introduction, the convergence result given in Theorem 2.2 also allows us to study the convergence (11) in the following Lebesgue spaces.
Corollary 2.1
With the same hypothesis of Theorem 2.2, for all \(1\le p \le +\infty \), \(1<q<+\infty \) and for all \(1+\varepsilon< \alpha < 2 \) we have the estimate:
We observe that in the framework of \(L^{p}_{t}L^{q}_{x}-\)spaces, the convergence rate is only driven by the parameter q, which describes the decaying properties of solutions in the spatial variable. Moreover, by setting the parameter \(\gamma =1\) and with the particular values \(p=q=2\), we obtain the following convergence rate: \( \Vert u_\alpha - u_2 \Vert _{L^{2}_{t}L^{2}_{x}} \lesssim (2-\alpha )^{1/2}\), which is similar to the one experimentally obtained in [2] for the time independent version of the equation (3): \((-\Delta )^{\alpha /2} u_\alpha =f_\alpha \).
To close this section, let us make the following final comments. First note that in this work we have restricted ourselves in the case when the parameter \(\alpha \) verifies \(1<\alpha <2\), however, our results are also valid for the case \(\alpha >2\) with minor technical modifications.
The lower constraint \(1+\varepsilon < \alpha \) (with \(0<\varepsilon \ll 1\)) given in Theorem 2.2 is essentially technical, due to estimates involving the expression \(\frac{1}{1-1/\alpha }\) (see, for instance, the estimate in Proposition 4.2 below). Consequently, our result left open the convergence problem when \(\alpha \rightarrow 1^{+}\) which is also interesting and could be a matter of further research.
Finally, we think that our method explained above could be also adapted to study the convergence given in (11) within the framework of other functional spaces, provided that we assume some natural hypothesis on the initial data.
Organization of the paper. In Sect. 3 we recall some well-known facts on the linear fractional heat equation that we will use in the next sections. Section 4 is devoted to the proof of Theorem 2.1, while in Sect. 5 we give a proof of Theorem 2.2 and Corollary 2.1.
3 Some well-known facts
In this section, for the completeness of this paper, we summarize some well-known facts on the linear and homogeneous fractional heat equation:
The fundamental solution of this equation, denoted by \(p_\alpha (t,x)\), can be computed via the Fourier transform by
Moreover, in the spatial variable the fundamental solution \(p_\alpha \) is given by
where the function \(P_\alpha \) is the inverse Fourier transform of \(e^{-\vert \xi \vert ^\alpha }\). See [13, Chapter 13] for more details. It is well-known that for \(1<\alpha <2\) the functions \(P_\alpha \) are smooth and positive. In addition, they verify the following pointwise inequalities
for a constant \(C>0\) and for all \(x\in \mathbb {R}^n\). These inequalities allow us to derive the following estimates.
Proposition 3.1
(\(L^p-\)estimates) For \(1 \le p \le +\infty \), there exists a constant \(C_{n,p} > 0\), which depends on the dimension \(n \in \mathbb {N}^*\) and the parameter p, such that for every \(1< \alpha < 2\) and for every \(t > 0\), we have
-
1.
\(\Vert p_{\alpha }(t, \cdot )\Vert _{L^p} \le C_{n,p} \ t^{-\frac{n}{\alpha }\left( 1-\frac{1}{p}\right) }\),
-
2.
\(\Vert \nabla p_{\alpha }(t, \cdot )\Vert _{L^p} \le C_{n,p} \ t^{-\frac{1+n\left( 1-\frac{1}{p}\right) }{\alpha }}\).
Moreover we have:
Proposition 3.2
(\(L^p-\)continuity) Let \(1 \le p < +\infty \). For every \(\varphi \in L^p(\mathbb {R}^n)\), we have \(\lim _{t \rightarrow 0^{+}}\Vert p_{\alpha }(t, \cdot )*\varphi - \varphi \Vert _{L^p}=0\).
On the other hand, by the identity \(\displaystyle {\widehat{p_\alpha }(t,\xi )=e^{-t\, \vert \xi \vert ^{\alpha }}}\) we have the following known results in the setting of the Sobolev spaces:
Proposition 3.3
(\(\dot{H}^s\) and \(H^s\) estimates) Let \(s_1, s_2 \ge 0\). Then, there is a constant \(C_{n,s_2} > 0\), which depends on the dimension \(n \in \mathbb {N}^*\) and the parameter \(s_2\), such that for every \(1 < \alpha \le 2\) and for every \(t > 0\), we have:
-
1.
\(\displaystyle {\Vert p_{\alpha }(t, \cdot )*\varphi \Vert _{\dot{H}^{s_1 +s_2}} \le C_{n,s_2} \ t^{-\frac{s_2}{\alpha }} \Vert \varphi \Vert _{\dot{H}^{s_1}}.}\)
-
2.
\(\displaystyle {\Vert p_\alpha (t, \cdot )*\varphi \Vert _{H^{s_1 + s_2}} \le C_{n,s_2} \left( 1 + t^{-\frac{s_2}{\alpha }}\right) \Vert \varphi \Vert _{H^{s_1}}.}\)
Proof
In order to verify the first point, we just write:
To verify the second point, let us start by writing
Then, for the first term on the right-hand side, by the Young’s inequalities and the point 1 in Proposition 3.1, we obtain
while for the second term on the right-hand side, by the point 1 proven above we can write:
Thus, the desired estimate follows directly from (17) and (18). \(\square \)
Proposition 3.4
(\(H^s-\) and \(\dot{H}^s-\)continuity) Let \(s_1, s_2 \ge 0\) and \(\varepsilon > 0\). There exists a constant \(C_{n,s_2, \varepsilon } > 0\), which depends on the dimension \(n \in \mathbb {N}^*\), the parameters \(s_2\) and \(\varepsilon \), such that for every \(1< \alpha < 2\) and for every \(t_1, t_2 > \varepsilon \), we have
-
1.
\(\Vert p_{\alpha }(t_1, \cdot )*\varphi - p_{\alpha }(t_2, \cdot )*\varphi \Vert _{\dot{H}^{s_1 +s_2}} \le C_{n,s_2,\varepsilon } \vert t_1 - t_2\vert ^{1/2} \vert \Vert \varphi \Vert _{\dot{H}^{s_1}}\),
-
2.
\(\Vert p_{\alpha }(t_1, \cdot )*\varphi - p_{\alpha }(t_2, \cdot )*\varphi \Vert _{H^{s_1 +s_2}} \le C_{n,s_2,\varepsilon } \vert t_1 - t_2 \vert ^{1/2} \Vert \varphi \Vert _{H^{s_1}}\).
Proof
To verify the first point, without loss of generality we shall assume that \( t_1>t_2>\varepsilon \). Then we write
We study now the expression \(\displaystyle {\vert e^{-(t_1-t_2) \vert \xi \vert ^\alpha } - 1\vert ^2}\). First, we remark that since we have \(t_1>t_2\) then the expression \(\displaystyle {\vert e^{-(t_1-t_2) \vert \xi \vert ^\alpha } - 1\vert }\) is uniformly bounded and we can write
Now, by the mean value theorem in the temporal variable we have
Thus, gathering these estimates we get
Getting back to the last integral we finally have:
hence, the first point is verified. The second point essentially follows these sames lines. \(\square \)
4 Global well-posedness and regularity: proof of Theorem 2.1
Let \(1 < \alpha \le 2\) fixed, and let \(u_{0} \in L^1\cap H^s(\mathbb {R}^n)\) be an initial datum. The result stated in Theorem 2.1 is well-known for the case \(\alpha =2\), see for instance [3, 8] and [9]. Consequently, we just consider the range \(1<\alpha <2\). As mentioned, the proof of this theorem is rather standard but, for the reader’s convenience, we shall detail some technical estimates.
Step 1: Local well-posedness. We consider the (equivalent) mild formulation given in (14), where the nonlinear term defines a multi-linear form in the variable u (see the expression (22) below). In order to construct a solution of the equation (14) we will use the Picard’s contraction principle for a time \(0<T<+\infty \) small enough. We thus consider the Banach space
endowed with the norm
Then, we will prove the following::
Proposition 4.1
Let \(s>n/2\) and let \(u_{0} \in L^1\cap H^s(\mathbb {R}^n)\) be an initial datum. Moreover, let \(1<\alpha <2\). Then, there exists a time given by:
where \(c>0\) is a numerical constant, and moreover, there exists a function \(u_\alpha \in E_T\) which is a solution of the equation (14).
Proof
We start by estimating the linear term in the equation (14).
Lemma 4.1
Let \(p_\alpha (t,x)\) be the kernel given in (16). Then we have \(\Vert p_\alpha (t,\cdot )*u_{0,\alpha } \Vert _{E_T} \le c\, (\Vert u_{0} \Vert _{L^1}+\Vert u_{0} \Vert _{H^s})\).
Proof
We first observe that, due to Proposition 3.2 and the first point in Proposition 3.3, the quantities \(\Vert p_\alpha (t,\cdot )*u_{0} \Vert _{L^1}\) and \(\Vert p_\alpha (t,\cdot )*u_{0} \Vert _{H^s}\) are continuous in the temporal variable.
On the other hand, by the Young’s inequalities and the point 1 in Proposition 3.1 (with \(p=1\)) we write
We also write
to obtain the wished estimate. \(\square \)
We study now the nonlinear term in the equation (14). For \(b\in \mathbb {N}\) with \(b\ge 2\), we denote the multi-linear form
where, to simplify our writing, we have written the function u instead of \(u_\alpha \). Then, we have the following estimate
Lemma 4.2
For \(u\in E_T\) we have \(M_b(u) \in E_T\). Moreover, the following estimate holds:
Proof
By [3] we have \(M_b(u)\in \mathcal {C}\big ([0,T], L^1(\mathbb {R}^n)\big )\), so it remains to prove that \(M_b(u) \in \mathcal {C}\big ([0,T], H^s(\mathbb {R}^n)\big )\). Indeed, let \(t_1, t_2 > 0\) and without loss of generality we assume that \(0< t_1 < t_2 \le T\). Then we write
For the first term on the right-hand side, by the point 2 in Proposition 3.4 (with \(s_1=s\) and \(s_2=1\)) and as \(s>n/2\), by the product laws in the Sobolev spaces we obtain:
Hence, we have \(\displaystyle {\lim _{t_1 \rightarrow t_2} R_{\alpha ,1}(t_1,t_2)=0}\). On the other hand, for the second term on the right-hand side we write
In order to estimate the term \(R_{\alpha ,2,1}(t_1,t_2)\), by the Hölder inequalities, the second point in Proposition 3.1, and moreover, the product laws in the Sobolev spaces, we write:
In addition, in order to estimate \(R_{\alpha ,2,2}(t_1,t_2)\), by Proposition 3.3 (with \(s_1=s\) and \(s_2=1\)), and by using again the product laws in the Sobolev spaces, we can write
By gathering the estimates made for the terms \(R_{\alpha ,2,1}(t_1,t_2)\) and \(R_{\alpha ,2,2}(t_1,t_2)\), we obtain \(\displaystyle {\lim _{t_1 \rightarrow t_2} R_{\alpha ,2}(t_1,t_2)=0}\). We thus have \(M_b(u)\in \mathcal {C}((0,T],H^s(\mathbb {R}^n))\). Now, we must prove the continuity at \(t=0\). For this we will verify the estimate
Indeed, by the Young’s inequalities, the second point in Proposition 3.1, Proposition 3.3, and moreover, the product laws in the Sobolev spaces we can write:
Once we have \(M_b(u) \in E_T\), we verify now the estimate stated in Lemma 4.2. First note that by the estimate (24) we can write
On the other hand, by applying the Young inequalities and the point 2 in Proposition 3.1 we have
Since \(s>n/2\) we have the embedding \(H^s(\mathbb {R}^n)\subset L^{\infty }(\mathbb {R}^n)\), and thus we can write
Then, we have
Finally, by (25) and (26) we obtain the desired estimate. This lemma is proven. \(\square \)
Once we have Lemmas 4.1 and 4.2 at our disposal, the rest of the proof of Proposition 4.1 follows from standard arguments. \(\square \).
Step 2: Regularity. We define the space \(H^{\infty }(\mathbb {R}^n)\) as \(\displaystyle {H^{\infty }(\mathbb {R}^n) = \bigcap _{s \ge 0} H^{s}(\mathbb {R}^n)}\).
Proposition 4.2
Let \(u_\alpha \in E_{T}\) be the unique solution of the equation (14) given by Proposition 4.1. This solution satisfies \(u_\alpha \in \mathcal {C}\big ((0,T],H^{\infty }(\mathbb {R}^n)\big )\). Moreover, we have \(u_\alpha \in \mathcal {C}^{1}((0,T], \mathcal {C}^{\infty }(\mathbb {R}^n))\); and for \(0<t\le T\) the solution \(u_\alpha \) verifies the differential equation (7) in the classical sense.
Proof
We will verify that each term on the right-hand side in the equation (14) belongs to the space \(\mathcal {C}\big ([0,T],H^{\infty }(\mathbb {R}^n)\big )\). For the first (linear) term, by the second point in Proposition 3.3, and moreover, by the second point in Proposition 3.4, we directly have \(\displaystyle {p_\alpha *u_{0,\alpha }\in \mathcal {C}\big ((0,T],H^{\infty }(\mathbb {R}^n)\big )}\).
For the second (nonlinear) term, we recall that by (24) for all time \(0<t\le T\) we have \(\int _{0}^{t} p_\alpha (t-\tau , \cdot )*\eta \cdot \nabla (u^{b}_{\alpha })(\tau , \cdot )d \tau \in H^s(\mathbb {R}^n)\). Then, we will prove that for \(\sigma > 0\) small enough we also have: \(\int _{0}^{t} p_\alpha (t-\tau , \cdot )*\eta \cdot \nabla (u^{b}_{\alpha })(\tau , \cdot )d \tau \in H^{s+\sigma }(\mathbb {R}^n)\). Indeed, by using the second point in Proposition 3.3, for \(\sigma >0\) (which we shall set later) we write
We thus set \(0<\sigma < \alpha -1\) (recall that we have \(1<\alpha < 2\)) to obtain that the last integral above computes down as
Then, for all time \(0<t\le T\) we obtain the estimate:
We will show now that we have
Let \(0< t_1, t_2 < T\), where, always without loss of generality we shall assume that \(t_1 < t_2\). Then we write:
where, we must study the terms \(\tilde{R}_{\alpha ,1}(t_1,t_2)\) and \(\tilde{R}_{\alpha ,2}(t_1,t_2)\). For the term \(\tilde{R}_{\alpha ,1}(t_1,t_2)\), by the second point in Proposition 3.3 we can write:
Since \(0<\sigma < \alpha -1\) the integral above computes down as
Hence, we have:
For the term \(\tilde{R}_{\alpha ,2}(t_1,t_2)\), always by the second point in Proposition 3.4, we can write:
Therefore, for \(0< \sigma < \alpha -1\), by (28) and (29) we have
At this point, we have proven that \(\displaystyle {u_\alpha \in \mathcal {C}((0, T_0], H^{s+\delta }(\mathbb {R}^n))}\) and by repeating this process (in order to obtain a gain of regularity for the nonlinear term) we conclude that \(\displaystyle {u_\alpha \in \mathcal {C}((0, T], H^{\infty }(\mathbb {R}^n))}\).
With this information at our disposal, we can verify now that for all \(0<t\le T\) and for all multi-index \(\textbf{a} \in \mathbb {N}^n\) we have \(\partial ^\textbf{a}_{x}u_\alpha (t,\cdot ) \in \mathcal {C}((0,T], \mathcal {C} \cap L^{\infty }(\mathbb {R}^n))\). Indeed, let \(\textbf{a}=(a_1,\cdots , a_n) \in \mathbb {N}^n\) be a multi-index, where we denote by \(\vert \textbf{a} \vert = a_1 + \cdots + a_n\) its size. Then, for \(\frac{n}{2}< s_1 < \frac{n}{2}+1\) we set \(s= \vert \textbf{a} \vert + s_1\). Since we have \(\displaystyle {u_\alpha \in \mathcal {C}((0, T], H^{\infty }(\mathbb {R}^n))}\) then we get \(\displaystyle {\partial ^{\textbf{a}}_{x} u_\alpha (t,\cdot ) \in H^{s_1}}(\mathbb {R}^n)\). Moreover, since \(\frac{n}{2}<s_1\) we have the continuous embedding \(H^{s_1}(\mathbb {R}^n)\subset L^{\infty }(\mathbb {R}^n)\), hence we conclude that \(\partial ^{\textbf{a}}_{x} u_\alpha (t,\cdot ) \in L^{\infty }(\mathbb {R}^n)\).
On the other hand, we recall that we have the identification \(H^{s_1}(\mathbb {R}^n)= B^{s_1}_{2,2}(\mathbb {R}^n)\) (where \(B^{s_1}_{2,2}(\mathbb {R}^n)\) denotes a non-homogeneous Besov space [1]). Moreover, we also have the continuous embedding \(B^{s_1}_{2,2}(\mathbb {R}^n)\subset B^{s_1 - n/1}_{\infty , \infty }(\mathbb {R}^n)\subset \dot{B}^{s_1-n/2}_{\infty ,\infty }(\mathbb {R}^n)\).
We thus have \(\displaystyle {\partial ^{\textbf{a}}_{x} u_\alpha (t,\cdot ) \in \dot{B}^{s_1-n/2}_{\infty ,\infty }(\mathbb {R}^n)}\). But, since \(\frac{n}{2}< s_1 < \frac{n}{2}+1\) then we have \(0< s_1 -\frac{n}{2}<1\), and thereafter, by definition of the homogeneous Besov space \(\dot{B}^{s_1-n/2}_{\infty ,\infty }(\mathbb {R}^n)\) (see always [1]) we get that \(\partial ^{\textbf{a}}_{x} u_\alpha (t,\cdot )\) is a \(\beta -\)Hölder continuous functions with parameter \(\beta = s_1 - \frac{n}{2}\in (0,1)\).
We have proven that \(u_\alpha \in \mathcal {C}((0,T], \mathcal {C}^{\infty }(\mathbb {R}^n))\). and we write \(\partial _t u_\alpha = - (-\Delta )^{\alpha / 2} u_\alpha - \eta \cdot \nabla (u^{b}_{\alpha })\) to obtain that \(\partial _t u_\alpha \in \mathcal {C}((0,T], \mathcal {C}^{\infty }(\mathbb {R}^n))\). Finally, we conclude that \(u_\alpha \in \mathcal {C}^{1}((0, T], \mathcal {C}^{\infty }(\mathbb {R}^n))\). Proposition 4.2 is proven. \(\square \)
Step 4: Global in time existence. By following similar arguments of [7] (see the proof of Theorem 2, page 9) we have the following result.
Proposition 4.3
Let \(u_0 \in L^1\cap H^s(\mathbb {R}^n)\) be an initial data and let \(T^* > 0\) be the maximal time of existence of the unique corresponding arising solution \(u_\alpha \in E_{T^*}\) (given by Proposition 4.1) to the problem (14). Then we have \(T^* = +\infty \).
Proof
Let us briefly explain the general idea of the proof. We assume that \(T^* < +\infty \). Then we will extend the solution \(u_\alpha \) beyond the time \(T^{*}\) to obtain a contradiction. We thus conclude that \(T^* = +\infty \).
We start by defining the following function \(T(\cdot ): [0, +\infty [ \rightarrow [0, +\infty [\) such that for each initial datum \(w_{0} \in L^1 \cap H^s(\mathbb {R}^n)\) the quantity \(T\left( \left\| w_{0} \right\| _{L^1}\right) \) is given by the expression
We recall that \(T\left( \left\| w_{0} \right\| _{L^1}\right) \) is precisely the first time of the existence of the solution \(w_\alpha \) to the equation (14), which is given by Proposition 4.1. Additionally, the key remark is that this function is decreasing in the variable \(\left\| w_{0} \right\| _{L^1}\).
On the other hand, by [3] we known that for every initial datum \(w_{0} \in L^1 \cap H^s(\mathbb {R}^n)\) we have a unique solution \(w_\alpha \in \mathcal {C}\big ([0,+\infty [,L^1(\mathbb {R}^n)\big )\) of the equation (14). Moreover, for every time \(t > 0\) we have the estimate
Since the function \(T(\cdot )\) defined above is decreasing in the variable \(\left\| w_{0} \right\| _{L^1}\), for the initial datum \(u_{0} \in L^1\cap H^s(\mathbb {R}^n)\) we can set a time \(0<T_1 < T^{*}\) such that for all \(w_{0} \in L^1\cap H^s(\mathbb {R}^n)\) with \(\left\| w_{0} \right\| _{L^1} \le \left\| u_{0} \right\| _{L^1}\) we have
Then, for \(0< \varepsilon < T_1\) small enough, we consider the time \(T^* - \varepsilon > 0\) and we set the initial datum \(w_0 = u_\alpha (T^* - \varepsilon , \cdot )\). We shall denote by \(w_\alpha \) its corresponding arising solution, which exists at least until the time \(T\big (\left\| w_{0} \right\| _{L^1}\big )\). We thus observe that the function
is a solution of the equation (14) associated to the initial datum \(u_{0}\). Moreover, we observe that this function is defined in the interval of time \(\big [0, T^* - \varepsilon +T\big (\left\| v_0 \right\| _{L^1}\big )\big ]\). But, by (30) we have \(\displaystyle { \left\| w_{0} \right\| _{L^1} = \left\| u_\alpha (T^* - \varepsilon , \cdot ) \right\| _{L^1} \le \left\| u_0 \right\| _{L^1}}\) and consequently by (31) we get \(T\big (\left\| w_{0} \right\| _{L^1}\big ) \ge T_1\).
Finally, we can write \( T^* - \varepsilon + T_1 \le T^* - \varepsilon + T\big (\left\| v_{0} \right\| _{L^1}\big )\); and since \(0< \varepsilon < T_1\) we obtain \(T^*<T^* - \varepsilon + T_1\), which is a contradiction with the definition of the time \(T^*\). Proposition 4.3 is proven. \(\square \)
Once we have proven the Propositios 4.1, 4.2 and 4.3, we can finish with the proof of Theorem 2.1. \(\square \)
5 From anomalous to classical diffusion
5.1 Proof of Theorem 2.2
For \(1<\alpha <2\), let \(u_\alpha \) be the mild solution of the equation (7) given by the expression (14). Moreover, for \(\alpha =2\) let \(u_2\) be the mild solution of the equation (7), which is given by the expression (15). Then, for a time \(0<T<+\infty \) fixed we write
where we must estimate each term on the right-hand side. For the term \(I_\alpha \) we write
In order to estimate the term \(I_{\alpha ,1}\), we apply the Bessel potential operators \((1-\Delta )^{-s/2}\) and \((1-\Delta )^{s/2}\) to obtain:
Then, by applying the Young inequalities (with \(1+1/\infty =1/2+1/2\)) we have
where we shall control each term above separately. For the first term on the right-hand side we have the following technical result:
Lemma 5.1
For \(s>n/2\) there exists a constant \(C=C(s)>0\) such that for all \(1<\alpha < 2\) we have:
Proof
First, we verify that the quantity \(\displaystyle { \left\| p_\alpha (t,\cdot )-h(t,\cdot ) \right\| ^{2}_{H^{-s}}}\) is continuous in the temporal variable t. Indeed, for \(0\le t_0, t\le T\) we have
As \(s>n/2\) we have \(\displaystyle {\int _{\mathbb {R}^n}\frac{d \xi }{(1+\vert \xi \vert ^2)^{s}} <+\infty }\); and then, we can apply the dominated convergence theorem to obtain that
Thereafter, by the continuity of the quantity \(\displaystyle { \left\| p_\alpha (t,\cdot )-h(t,\cdot ) \right\| ^{2}_{H^{-s}}}\) with respect to the variable t, there exists a time \(0 < t_1 \le T\) such that
Now, we will prove the estimate \(\displaystyle {\left\| p_\alpha (t_1,\cdot )-h(t_1,\cdot ) \right\| _{H^{-s}}\le C\, T\, \vert 2- \alpha \vert }\). For this we write:
Here, for \(\xi \in \mathbb {R}^n\setminus \{0\}\) fixed, and for \(1<\alpha < 2 +\delta \) (with \(\delta >0\)) we define the function
where, by computing its derivative with respect to the variable \(\alpha \) we get
Thus, by the mean value theorem (in the variable \(\alpha \)) we can write
Moreover, we can also prove the uniform estimate with respect to the variable \(\xi \):
The proof of this estimate is not difficult and it is given in detail at the Appendix 6. We thus have,
Then, we get back to the identity (35) and we can write
\(\square \)
On the other hand, to control the second term on the right-hand side in the expression (34), we recall that by hypothesis (10) the family \(\displaystyle {(u_{0,\alpha })_{1<\alpha <2}}\) is bounded in \(H^s(\mathbb {R}^n)\).
Thus, for the term \(I_{\alpha ,1}\) given in (33) we can write:
We study now the term \(I_{\alpha ,2}\), which is also given in (33). By the Young inequalities (with \(1+ 1/\infty = 1 + 1/ \infty )\), the well-known properties of the heat kernel, and moreover, by the assumption give in (12), we have:
Consequently, with the estimates (38) and (39) above we obtain:
We study now the term \(J_{\alpha }\) given in the expression (32). For this we write
where we will study the terms \(J_{\alpha ,1}\) and \(J_{\alpha ,2}\) separately. For the term \(J_{\alpha ,1}\), we apply first the operators \((1-\Delta )^{-s/2}\) and \((1-\Delta )^{s/2}\), and moreover, by the Young inequalities (with \(1+1/\infty = 1/2 + 1/2\)) we have
In order to control the first term on the right-hand side, we follow the same lines in the proof of Lemma 5.1 with the function \(\displaystyle {f_\xi (\alpha )= i \xi _j \, e^{- t_1 \vert \xi \vert ^\alpha }}\), with \(j=1,2,\cdots , n\). Then we have
Thereafter, to control the second term on the right-hand side we shall need the following:
Lemma 5.2
There exists \(0<\varepsilon \ll 1\), and there exists a constant
such that for all \(1+\varepsilon< \alpha < 2\) we have:
Proof
For any the initial data \(u_{0,\alpha } \in L^1 \cap H^s(\mathbb {R}^n)\), with \(1<\alpha <2\), we recall that by Proposition 4.1 there exists a time \(T_\alpha \) (depending on \(\alpha \)) defined by (21) as
and there exists a (unique) solution \(u_\alpha \in E_{T_\alpha }\) of the equation (14). Our staring point is to obtain a lower bound for the time \(T_\alpha \) which does not depend on \(\alpha \).
By our hypothesis given in (10) we can set \(0<\varepsilon \ll 1\) such that for all \(1+\varepsilon< \alpha <2\) we have:
hence we get the control
and then, we can write:
Moreover, since \(1+\varepsilon< \alpha <2\) then the expression on the left-hand can be estimated from below by the following quantity:
At the Appendix 7 we verify in detail this estimate. Then, for all \(1+\varepsilon<\alpha < 2\) we have \(T_0 \le T_\alpha \).
Once we have the lower estimate \(T_0\le T_\alpha \), we remark that for all \(1+\varepsilon< \alpha < 2\) the solution \(u_\alpha \) of the equation (14), which is constructed in the Proposition 4.1 by the Picard’s fixed point argument, verifies \(u_\alpha \in E_{T_0}\) and moreover we have the estimate \(\displaystyle {\Vert u_\alpha \Vert _{E_{T_0}} \le c_0(\Vert u_{0,\alpha } \Vert _{L^1}+\Vert u_{0,\alpha } \Vert _{H^s})}\).
We also remark that by Proposition 4.3 the solution \(u_\alpha \) is extended to a global in time solution by a well-known iterative argument: for every interval \([kT_0, (k+1)T_0]\) (with \(k \in \mathbb {N}^{*}\)) we set the initial initial datum \(u_\alpha (k T_0, \cdot )\) and we apply again the Picard’s fixed point schema to obtain a (unique) solution \(u_\alpha \in E_{[k T_0, (k+1)T_0]}\) (recall that the space \( E_{[k T_0, (k+1)T_0]}\) is defined in (19) and (20)). Moreover, there exists a constant \(c_k >0\) such that we have
We study now the expression \(\displaystyle {c_k (\Vert u_\alpha (k T_0, \cdot ) \Vert _{L^1}+\Vert u_\alpha (k T_0, \cdot ) \Vert _{H^s})}\). For the quantity \(\displaystyle {\Vert u_\alpha (k T_0, \cdot ) \Vert _{L^1} }\), by (30) we have \(\displaystyle {\Vert u_\alpha (kT_0, \cdot )\Vert _{L^1} \le \Vert u_{0,\alpha } \Vert _{L^1}}\). Then, we can write
On the other hand, for the quantity \(\displaystyle {\Vert u_\alpha (k T_0, \cdot ) \Vert _{H^s}}\), we remark that we have
Thus, we can iterate these estimates and for a constant \(C_k>0\) big enough (in particular we must have \(\displaystyle {C_k > \prod _{j=0}^{k} c_j}\)) we obtain \(\Vert u_\alpha (k T_0, \cdot ) \Vert _{H^s} \le C_k (\Vert u_{0,\alpha } \Vert _{L^1}+\Vert u_{0,\alpha } \Vert _{H^s})\). Consequently, for all \(k\in \mathbb {N}^{*}\) we have
Now, we are able to prove the estimate (44) stated in this lemma. For the time T there exists \(k_T \in \mathbb {N}\) (which depends on T) such that we have \(k_T T_0 \le T \le (k_T+1) T_0\). Then, as \(s>n/2\) by the product laws in the Sobolev spaces we can write
Then, by the control given in (47) and the control given in (45) we have
To finish the proof of this lemma, we just remark that the constant \(\mathcal {C}\) defined above also depends on the parameter \(\varepsilon \), since the \(k_T\) depends on T and \(T_0\); and the time T given in (46) depends on \(\varepsilon \). \(\square \)
With the estimates (43) and (44) at our disposal, we get back to the estimate (42) to write
On the other hand, We study now the term \(J_{\alpha ,2}\) given in (41). For this, by the Young inequalities we write:
Here, by the well-known properties of the heat kernel \(h(t,\cdot )\) we have the estimate \(\displaystyle {\Vert \nabla h(t-s,\cdot )\Vert _{L^1} \le c (t-s)^{-1/2}}\). Thereafter, in order to estimate the term \(\displaystyle {\Vert u^{b}_{\alpha }(s,\cdot )- u^{b}_{2}(s,\cdot ) \Vert _{L^\infty }}\), since \(s>n/2\) and by Lemma 5.1 we can write
where the constant \(C>0\) does not depend on \(\alpha \). With these estimates we obtain
and consequently we have
Once we estimated the terms \(I_\alpha \), \(J_{\alpha ,1}\) and \(J_{\alpha ,2}\) in (40), (48) and (49) respectively, we get back to (32) we can write
In this estimate, first we set a time \(0<T_1<T\) small enough such that it verifies
we get:
and then we can write
Hence, by (40) and (48) we obtain
Finally, we iterate this argument on the intervals \([kT_1, (k+1)T_1]\), with \(k \in \mathbb {N}\), and then, for the time \(0<T<+\infty \) we have
Theorem 2.2 is now proven. \(\square \)
5.2 Proof of Corollary 2.1
For \(0<T<+\infty \) fixed, and moreover, for \(1\le p < +\infty \) and \(1< q < +\infty \), by the interpolation inequalities in the spatial variable (with \(\theta = 1/q\)) we write
By the hypothesis (10), we can set \(M>0\) (small enough) such that for \(1+\varepsilon< \alpha <2\) we have \(\displaystyle {\Vert u_\alpha (t,\cdot )-u_2(t,\cdot )\Vert _{L^1} \le M}\). Then we obtain
Thus, the wished estimate follows from the inequality (13) proven in Theorem 2.2. Moreover, the case \(p=+\infty \) follows the same lines above with the obvious modifications. \(\square \)
Data Availibility
Data sharing not applicable to this article as no datasets were generated or analyzed during the current study.
References
Bahouri, H., Chemin J.Y., Danchin, R.: Fourier Analysis and nonlinear partial differential equations. Springer Vol: 343 (2011)
Biccari, U., Hernández-Santamaía, V.: The poisson equation from non-local to local. Electron. J. Differ. Equ. 2018(145), 1–13 (2018)
Biler, P., Karch, G., Woyczynski, W.A.: Asymptotics for conservation laws involving Lévy diffusion generators. Studia Math. 148, 171–192 (2001)
Biler, P., Funaki, T., Woyczynski, W.A.: Fractal burgers equations. J. Differ. Equ. 148(1), 9–46 (1998)
Biler, P., Karck, G., Woyczyński, W.: Asymptotics for conservation laws involving Lévy diffusion generators. Studia Mathematica 148(2), 171–192 (2001)
Brandolese, L., Karch, G.: Far field asymptotics of solutions to convection equation with anomalous diffusion. J. Evol. Equ. 8, 307–326 (2008)
Cui, S., Zhao, X.: Well-posedness of the Cauchy problem for Ostrovsky, Stepanyams and Tsimring equation with low regularity data. J. Math. Anal. Appl. 344, 778–787 (2008)
Droniou, J., Gallouët, T., Vovelle, J.: Global solution and smoothing effect for a non-local regularization of a hyperbolic equation. J. Evol. Eq. 3, 499–521 (2002)
Droniou, J., Imbert, C.: Fractal first order partial differential equations. Arch. Rat. Mech. Anal. 182, 299–331 (2006)
Funaki, T., Surgailis, D., Woyczynski, W.A.: Gibbs-Cox random fields and Burgers turbulence. Ann. Appl. Prob. 5, 701–735 (1995)
Funaki, T., Woyczynski, W. A.: Interacting particle approximation for fractal Burgers equation. Stochastic processes and related topics, a volume in memory of stamatis cambanis, Birkhaüser, Boston (1998)
Ignat, L., Rossi, J.D.: A non-local convection-diffusion equation. J. Funct. Anal. 251(2), 399–437 (2007)
Jacob, N.: Pseudo-differential operators and Markov processes, vol. I. Fourier analysis and semi-groups, Imperial College Press, London (2001)
Jourdain, B., Méléard, S., Woyczynski, W.A.: A probabilistic approach for nonlinear equations involving the fractional Laplacian and singular operator. Potential Anal. 23, 55–81 (2005)
Saichev, A.S., Woyczynski, W.A.: Advection of passive and reactive tracers in multi-dimensional Burgers velocity field. Physica D 100, 119–141 (1997)
Saichev, A. S., Woyczynski, W. A.: Distributions in the physical and engineering sciences. Distributional and fractal calculus integral transforms and wavelets, Vol. 1, Birkhauser, Boston (1997)
Shlesinger, M.F., Zaslavsky, G.M., Frisch, U.: Lévy flights and related topics in physics. Lecture notes in physics, vol. 450. Springer, Verlag Berlin (1995)
Vazquez, J.L., de Pablo, A., Quiros, F., Rodriguez, A.: Classical solutions and higher regularity for nonlinear fractional diffusion equations. J. Eur. Math. Soc. 19(7), 1949–1975 (2017)
Zaslavsky, G.M.: Fractional kinetic equations for Hamiltonian chaos. Physica D 76, 110–122 (1994)
Zaslavsky, G.M., Abdullaev, S.S.: Scaling properties and anomalous transport of particles inside the stochastic layer. Phys. Rev. E 51(5), 3901–3910 (1995)
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflicts of interest
This work has not received any financial support. In addition, the authors declare that they have no conflicts of interest.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix
We prove here the estimate (37). We recall the expression
Then, we write
where, we shall estimate the terms A and B separately. For the term A, as we have \(\vert \xi \vert \le 1\), \(1<\alpha <2+\delta \), and moreover, as we have \(\displaystyle {\lim _{\vert \xi \vert \rightarrow 0^{+}} \vert \xi \vert \ln (\vert \xi \vert )=0}\), then we can write:
For the term B, since \(\vert \xi \vert >1\) then we can write
Appendix
Here we give a proof of the estimate
First, as we have \(1+\varepsilon< \alpha < 2\), then we get \(1-\frac{1}{1+\varepsilon }< 1-\frac{1}{\alpha }\), and we can write
Thereafter, by the sake of simplicity, we denote
and we have
We study now the expression \(\frac{\alpha }{\alpha -1}\). Since we have \(1+\varepsilon< \alpha <2\) then we get \(1+\varepsilon< \frac{\alpha }{\alpha -1} < \frac{2}{\varepsilon }\). Thus, on the one hand, if the quantity (a) above verifies \((a)<1\) then we have \(\displaystyle {\frac{1}{2}[(a)]^{\frac{2}{\varepsilon }} \le \frac{1}{2}[(a)]^{\frac{\alpha }{\alpha -1}}}\). On the other hand, if the quantity (a) verifies \((a)\ge 1\) then we have \(\displaystyle {\frac{1}{2}[(a)]^{1+\varepsilon } \le \frac{1}{2}[(a)]^{\frac{\alpha }{\alpha -1}}}\).
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Jarrín, O., Loachamín, G. From anomalous to classical diffusion in a nonlinear heat equation. J Elliptic Parabol Equ 9, 1071–1099 (2023). https://doi.org/10.1007/s41808-023-00237-5
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s41808-023-00237-5
Keywords
- Nonlinear heat equation
- Fractional Laplacian operator
- Asymptotic behavior of solutions depending on the diffusion parameter
- From non local to local behavior