Abstract
In his celebrated article, Aronson established Gaussian bounds for the fundamental solution to the Cauchy problem governed by a second order divergence form operator with uniformly elliptic coefficients. We extend Aronson’s proof of upper heat kernel estimates to nonlocal operators whose jumping kernel satisfies a pointwise upper bound and whose energy form is coercive. A detailed proof is given in the Euclidean space and extensions to doubling metric measure spaces are discussed.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
1.1 Background
Let \(a_{i,j} : [0,T] \times \mathbb {R}^d \rightarrow (0,\infty )\), \(i,j = 1,\dots ,d\), be bounded, measurable coefficients which satisfy the usual uniform ellipticity condition. A celebrated result by D.G. Aronson from 1967 says that the fundamental solution \(\Gamma (y,s;x,\eta )\) to the equation
satisfies the following two-sided estimate for all \(0 \le \eta< s < T\), and \(x,y \in \mathbb {R}^d\):
where \(c_1,c_2,c_3,c_4 > 0\) depend only on d and the ellipticity constants. In other words, the fundamental solution of the classical heat equation \(\partial _t u - \Delta u = 0\) is an upper and lower bound of \(\Gamma \) up to multiplicative constants, see [1, 2]. In this sense, the bounds (1.2) are robust in the class of second order divergence form operators with bounded, measurable, uniformly elliptic coefficients.
Aronson’s proof is closely related to the so-called DeGiorgi-Nash-Moser theory for parabolic differential operators of second order with bounded, measurable and uniformly elliptic coefficients. The proof heavily relies on Hölder regularity estimates and the parabolic Harnack inequality for solutions to (1.1), see also [3].
[2] has initiated several research studies on estimates for fundamental solutions to parabolic equations in various contexts. An important feature of this research is that it connects partial differential equations with geometry. This is due to the sensitivity of the heat kernel to the geometric properties of the underlying space. This phenomenon becomes apparent in the celebrated works [18, 41], where the method of Aronson was generalized to prove heat kernel estimates on complete Riemannian manifolds with nonnegative Ricci curvature. Some of their arguments have been further refined and generalized in [20], where an integral estimate for the heat kernel was established that is useful for proving the upper bound in (1.2). We refer the interested reader to [32, 39, 50] and the references therein for more detailed expositions on this topic.
1.2 Main results
The goal of this article is to extend Aronson’s proof of upper heat kernel estimates to integro-differential operators of the form
Such operators are determined by a jumping kernel \(k : (0,T) \times \mathbb {R}^d \times \mathbb {R}^d \rightarrow [0,\infty ]\) which is assumed to be symmetric, i.e., \(k(t;x,y) = k(t;y,x)\) and satisfies a pointwise upper bound
for some given constant \(\Lambda > 0\), and \(\alpha \in (0,2)\), \(0 < T \le \infty \).
Moreover, we assume that there is \(\lambda > 0\) such that for any ball \(B \subset \mathbb {R}^d\) and every \(v \in H^{\alpha /2}(B)\):
\(\mathcal {E}_{\ge }\) can be thought of as a coercivity assumption on k and is substantially weaker than a pointwise lower bound. We refer the reader to Sect. 2 for a more detailed discussion and to Section4 where we explain a possible extension of our method and replace \(\mathcal {E}_{\ge }\) by a Faber-Krahn inequality.
We are now ready to state the main result of this article in the aforementioned setup. For a possible extension to doubling metric measure spaces and jumping kernels of mixed type, we refer to Theorem 4.1.
Theorem 1.1
Let \(k : (0,T) \times \mathbb {R}^d \times \mathbb {R}^d \rightarrow [0,\infty ]\) be symmetric and assume \(k_{\le }\), \(\mathcal {E}_{\ge }\). Let \(p(y,s;x,\eta )\) be the fundamental solution to the equation
where \(\eta \in [0,T)\). Then there exists a constant \(c > 0\) depending on \(d,\alpha ,\lambda ,\Lambda \) such that for every \(0 \le \eta< s < T\), and \(x,y \in \mathbb {R}^d\):
Estimate (1.4) states that the fundamental solution to \(\partial _t u - L_t u = 0\) possesses the same upper bound as the fundamental solution to the fractional heat equation \(\partial _t u + (-\Delta )^{\alpha /2} u = 0\), see [8]. Only for \(\alpha = 1\) an explicit formula for \(p(y,s;x,\eta )\) is known:
It is well known that corresponding lower bounds do not hold under \(k_{\le }\) and \(\mathcal {E}_{\ge }\) since the latter condition does not rule out k to be zero in certain cones of directions (see [34]).
Different versions of Theorem 1.1 already exist in the literature. Let us give a brief account on the history of heat kernel bounds for nonlocal operators. Two-sided estimates of the form (1.4) have been established by [11, 13] using a probabilistic approach. They assume that the jumping kernel is pointwise comparable to \(\vert x-y \vert ^{-d-\alpha }\), \(\alpha \in (0,2)\), from above and below. Their analysis of the upper heat kernel estimate heavily relies on [16], where Davies’ method was extended to a more general setup, including jump processes. Building upon this, [9] derived (1.4) assuming \(k_{\le }\) and a Nash inequality.
In a series of articles, see [28,29,30], the analysis of heat kernel estimates was extended to metric measure spaces with walk dimension greater than 2. The authors were able to characterize upper heat kernel estimates, as well as two-sided heat kernel estimates in terms of equivalent conditions on the jumping kernels and the geometry of the underlying space. Their approach does not use the underlying stochastic process and is based on certain comparison inequalities of the corresponding heat semigroups relying on the parabolic maximum principle. Note that Davies’ method was extended to jumping kernels with jump index \(\alpha >2\) in [33, 48]. The aforementioned results assume certain homogeneity of the doubling measure space and do not deal with mixed-type jumping kernels.
In [14, 15, 17] upper and two-sided heat kernel estimates were investigated on doubling metric measures spaces for jumping kernels of mixed type. This approach applies also to cases when \(\alpha \ge 2\). We would like to draw the reader’s attention to Theorem 1.15 in [17]. In the case \(\alpha < 2\), it states that upper heat kernel estimates of the form (1.4) are equivalent to a pointwise upper bound on the jumping kernel and a Faber-Krahn inequality, which can be understood as an implicit lower bound.
A major difference between our approach and [17] is that our method relies on purely analytic arguments, while [17] makes essential use of the corresponding stochastic process. In Theorem 4.1 we extend our approach to doubling metric measure spaces and jumping kernels of mixed type. Let us mention that we prove on-diagonal heat kernel estimates with the help of a parabolic \(L^\infty - L^1\)-estimate, see Lemma 4.2. This rather straightforward approach allows us to avoid truncation methods and the usage of the iteration techniques of [35].
In contrast to our setup, all jumping kernels in the results discussed above are assumed to be time-homogeneous. Note that it would require substantial effort to extend methods based on stochastic processes to situations with time-dependent jumping kernels. Heat kernel estimates for time-inhomogeneous jumping kernels were established in [43, 44] where the authors assume pointwise upper and lower bounds on the jumping kernel. Note that assuming pointwise lower bounds is more restrictive than \(\mathcal {E}_{\ge }\). The focus of these works lies on the treatment of an additional divergence-free drift of first order.
There are further results on heat kernel estimates for nonlocal operators, which are related to Theorem 1.1. For example, sharp two-sided estimates for jump processes on \(\mathbb {R}^d\) with upper scaling index not strictly less than 2 are established in [10]. In [36], heat kernel estimates for a certain class of jump processes with singular jumping measures are proved.
1.3 Strategy of proof
A main insight of Aronson’s proof for second order differential operators is the observation that solutions u to the Cauchy problem (1.1) satisfy the weighted \(L^2\)-estimate
for \(0 \le \eta< s < T\), whenever H satisfies
for a given number \(C > 0\) depending on the ellipticity constants. (1.6) is closely related to the famous Li-Yau inequality:
In fact, a direct computation reveals that the Gauss-Weierstrass kernel \(w(t,x) = t^{-\frac{d}{2}}e^{-\frac{\vert x \vert ^2}{4t}}\) satisfies (1.7) with equality. By a scaling argument, it becomes evident that (1.6) holds true for \(H(t,x) = (C[t])^{\frac{d}{2}} w(C[t],x-y) = \exp (-\frac{\vert x-y \vert ^2}{4 C [t]})\), where \([t] {:}{=} 2(s-\eta ) - (t-\eta )\) and \(y \in \mathbb {R}^d\) can be chosen arbitrarily.
This insight suggests that some qualitative information on the decay of solutions to (1.1) is encoded in the weighted \(L^2\)-estimate (1.5). Indeed, by combining (1.5) with a localized \(L^{\infty }-L^2\)-estimate, as it was proved by Moser [45,46,47], one can estimate the value of the solution to the Cauchy problem at the center of a ball that lies outside the support of the initial data (see Theorem 3.2). From such estimate, it is not difficult to deduce
for every \(\sigma > 0\) and \(0 \le \eta< s < T\) with \(s-\eta \le \sigma ^2\). Together with the on-diagonal estimate \(\Gamma (y,s;x,\eta ) \le c(s-\eta )^{-\frac{d}{2}}\), one deduces the upper bound in (1.2) via a standard argument.
We would like to point out that (1.7) is at the core of the works [18, 41], where (1.7) was used to derive a parabolic Harnack inequality on Riemannian manifolds with nonnegative Ricci-curvature and to establish its equivalence to Gaussian heat kernel bounds.
Next, let us summarize how we adapt Aronson’s proof to integro-differential operators. First, we require a nonlocal analog of (1.5). We prove that there exist functions H that satisfy
given \(C,\rho > 0\). Here, \(\Gamma ^{\alpha }_{\rho }\) denotes the \(\rho \)-truncated carré du champ operator of order \(\alpha \in (0,2)\), which is defined as follows:
By a careful choice of a function H satisfying (1.8), we deduce a nonlocal analog of (1.5) for solutions to the \(\rho \)-truncated Cauchy problem, see Lemma 3.1. The corresponding integro-differential operator only takes into account differences up to distance \(\rho \). In order to prove an a priori bound for the corresponding fundamental solution \(p_{\rho }\) (see Theorem 3.4), we derive a parabolic \(L^{\infty }-L^2\)-estimate in the spirit of [51], see Lemma 2.4. The estimate involves a nonlocal, truncated tail-term which requires special treatment. In a final step, we obtain the desired upper heat kernel estimate (1.4) for p by gluing together short and long jumps. Such argument is by now standard in the theory of jump processes.
Last, we explain several difficulties that occur when avoiding the detour via the truncated jumping kernel.
First, the corresponding \(L^{\infty }-L^2\)-estimate involves a non-truncated tail-term which cannot be controlled without any further assumptions on k.
Second, finding suitable weight functions H which satisfy a non-truncated version of (1.8) is a challenging task in the light of the following observation: The corresponding nonlocal analog of the Li-Yau inequality, which would imply an estimate of the form (1.5) for solutions to (1.3), reads as follows:
However, one can show that the fundamental solution \(w_{\alpha }(t,x)\) to \(\partial _t u + (-\Delta )^{\alpha /2}u = 0\) does not satisfy (1.9). Let us give a quick proof of this fact. First, note that \(w_{\alpha }\) is a radial function and satisfies
For a proof of this identity, we refer to (2.5) in [52]. Consequently,
but this is a contradiction to (1.9) since \(\Gamma ^{\alpha }(w_{\alpha }^{1/2},w_{\alpha }^{1/2})(t,0) > 0\) for \(t > 0\).
We would like to point out that modifications of the estimates (1.6) and (1.7) also hold true in the context of the porous medium equation. Such estimates are known as Aronson-Benilan estimates (see [4, 40]). For a discussion of nonlinear fractional diffusion equations of porous medium type, we refer the interested reader to [12, 24, 53].
1.4 Outline
This article is separated into five sections. In Sect. 2 we present several auxiliary results that we need in our proof. Section 3 contains the derivation of the upper heat kernel bounds and proves Theorem 1.1. In Sect. 4, we explain how our method can be applied to jumping kernels of mixed type on metric measure spaces. In Sect. 1, we provide a proof of a gluing lemma which differs from [27] due to the time-inhomogeneity of the jumping kernels under consideration.
2 Preliminaries
In this section we provide several auxiliary results that will be required for the proof of Theorem 1.1 in Sect. 3.
Let \(k : (0,T) \times \mathbb {R}^d \times \mathbb {R}^d \rightarrow [0,\infty ]\) be a symmetric jumping kernel satisfying the pointwise upper bound \(k_{\le }\) and the coercivity condition \(\mathcal {E}_{\ge }\).
Let us make a few comments on assumption \(\mathcal {E}_{\ge }\): First of all, \(\mathcal {E}_{\ge }\) can be regarded as a nonlocal substitute of the classical uniform ellipticity condition for local operators. In fact, it is considerably weaker than a pointwise lower bound on the jumping kernel, since \(\mathcal {E}_{\ge }\) allows for jumping kernels that might degenerate in certain directions, as for example kernels that are supported on double cones. We refer the interested reader to [19, 21] for an investigation of such condition. Since k is allowed to depend on time, our result also covers kernels that are supported on double cones with fixed apex, whose cone axes rotate continuously in time.
A coercivity assumption like \(\mathcal {E}_{\ge }\) is crucial to our approach since it is needed for the \(L^{\infty }-L^2\)-estimate (Lemma 2.4) and also the on-diagonal heat kernel bound (Theorem 2.3). In the literature, lower bounds on jumping kernels are often introduced through functional inequalities, e.g. in [9], or [17] where the authors assume a Nash -, or a Faber-Krahn inequality. We point out that such assumption would have been possible also in this work, since the proofs of Lemma 2.4, Theorem 2.3 can be changed accordingly (see Sect. 4). For a discussion on the equivalence of Nash - and Faber-Krahn inequalities, we refer the reader to [17]. Moreover, we would like to mention the recent article [7], where the relation between \(L^1-L^{\infty }\) smoothing effects, on-diagonal upper heat kernel estimates, and functional inequalities are studied for fractional equations of porous medium type.
Let us point out a possible extension that was mentioned to us by a reviewer of this article. Our approach certainly allows to track constants in the respective estimates, as it is done in [6]. In particular, one could treat time-dependent versions of the condition \(\mathcal {E}_{\ge }\) and establish sufficient conditions on \(\lambda (t)\) as \(t \rightarrow \infty \).
For any \(\rho > 0\), we define the truncated jumping kernel \(k_{\rho }\) via
The associated integro-differential operator \(L_t^{\rho }\) is defined as
Definition 2.1
We say that a function \(u \in L^2_{loc}((\eta ,T);H^{\alpha /2}(\mathbb {R}^d))\) with \(\partial _t u \in L^1_{loc}((\eta ,T) ; L^2_{loc}(\mathbb {R}^d))\) solves the Cauchy problem associated with k in \((\eta ,T) \times \mathbb {R}^d\):
if for every \(\phi \in H^{\alpha /2}(\mathbb {R}^d)\) with \({{\,\textrm{supp}\,}}(\phi )\) compact, it holds
where we write
for the family of energy forms \((\mathcal {E}_t)_{t \in (\eta ,T)}\) associated with k.
Solutions to the \(\rho \)-truncated Cauchy problem associated with k in \((\eta ,T) \times \mathbb {R}^d\)
are defined in an analogous way, replacing k by \(k_{\rho }\).
Throughout this article, we will assume that the fundamental solutions \(p, p_{\rho } : (0,T) \times \mathbb {R}^d \times [0,T) \times \mathbb {R}^d \rightarrow [0,\infty ]\) to the equations \(\partial _t u - L_t u = 0\) and \(\partial _t u - L^{\rho }_t u = 0\) exist. p and \(p_{\rho }\) have the following properties for all \(0 \le \eta< t< s < T\), \(x,y \in \mathbb {R}^d\):
Moreover, the solutions u to (2.1) and \(u_{\rho }\) to (2.5) are unique and have the representation
In the following, we will denote the unique solutions to (2.1) and (2.5) by \(P_{\eta ,s}u_0\), and \(P^{\rho }_{\eta ,s}u_0\). \((P_{\eta ,s})_{s \in [\eta ,T)}\), and \((P^{\rho }_{\eta ,s})_{s \in [\eta ,T)}\) are called the heat semigroups associated with k, and \(k_{\rho }\).
In the time-homogeneous case, i.e., when k does not depend on t, the existence of \((P_{\eta ,s})_{s \in [\eta ,T)}\) and \((P^{\rho }_{\eta ,s})_{s \in [\eta ,T)}\) is guaranteed by symmetric Dirichlet form theory. The existence of the fundamental solution classically follows from so-called ultracontractivity estimates for the heat semigroup which are a consequence of Nash’s inequality. For time-inhomogeneous jumping kernels k which satisfy the following pointwise lower bound for some \(\lambda > 0\)
the existence of the fundamental solutions p and \(p_{\rho }\) was proved in [43] by approximation of k through a sequence of smooth jumping kernels for which the desired properties follow from the theory of pseudo-differential operators. A similar result is proved in [37, 38] but under an additional smoothness assumption on t. Alternatively, the existence and uniqueness of solutions to (2.1) and (2.5) can be established by following the proof of Theorem 5.3 in [25], which is based on a parabolic version of the Lax-Milgram lemma (see Corollary 23.26 in [54]).
The following result explains the connection between p and \(p_{\rho }\) and is crucial to our approach. It is frequently used in the derivation of upper heat kernel bounds for alpha-stable like processes and goes back to a probabilistic construction carried out in [42]. An analytic proof via the parabolic maximum principle is derived in [27] (see also [30]). Since both proofs are known only in the time-homogeneous case, we will provide a modified version of the argument in [27] in the appendix.
Lemma 2.2
Assume that k satisfies \(k_{\le }\), \(\mathcal {E}_{\ge }\). Then there exists \(c > 0\) such that for every \(\rho > 0\), \(0 \le \eta< s < T\), and \(x,y \in \mathbb {R}^d\) it holds
Next, we provide the so-called on-diagonal bound for the heat kernels p, and \(p_{\rho }\).
Theorem 2.3
(on-diagonal bound) Assume that k satisfies \(k_{\le }\), \(\mathcal {E}_{\ge }\). Then there exists \(c > 0\) depending on \(d,\alpha ,\lambda ,\Lambda \) such that for every \(\rho > 0\), \(0 \le \eta< s < T\), and \(x,y \in \mathbb {R}^d\) it holds
There are at least two ways to prove Theorem 2.3. One approach classically goes via Nash inequalities (see [17]) and can be traced back to Nash’s famous work [49]. This proof also works in the time-inhomogeneous setup (see [43]).
Another way to establish on-diagonal bounds goes via \(L^{\infty }-L^1\)-estimates. For this, we refer to Lemma 4.2, where such estimate is proved in a more general setup. Observe that \((t,z) \mapsto p(y,t;z,\eta )\) solves \(\partial _t u - L u = 0\) in \((\eta ,T) \times \mathbb {R}^d\) for every \(y \in \mathbb {R}^d\). Therefore, by the \(L^{\infty }-L^1\)-estimate (4.4), for every \(0 \le \eta< s < T\) and \(x,y \in \mathbb {R}^d\):
where we used (2.7) in the last step. (2.14) follows from Lemma 2.2.
The remainder of this section is devoted to proving an \(L^{\infty }-L^2\)-estimate for solutions to the truncated problem \(\partial _t u - L^{\rho }_t u = 0\) in a time-space cylinder \(I_R(t_0) \times B_R(x_0)\), where \(t_0 \in (0,T)\), \(x_0 \in \mathbb {R}^d\), and \(I_R(t_0) {:}{=} (t_0 - R^{\alpha },t_0) \subset (\eta ,T)\).
For truncated jumping kernels \(k_{\rho }\), it is possible to estimate the nonlocal tail-term by an \(L^2\)-norm over a ball with radius \(\rho \), see (2.16), without assuming any pointwise lower bounds of k, or a UJS-type condition as in [5]. This is at the cost of the suboptimal scaling factor \(( \rho ^{\alpha }/R^{\alpha })^{d/(2\alpha )}\) in the resulting estimate, which luckily does not affect the proof of the final heat kernel estimate.
A similar result for solutions to elliptic equations was obtained in [17]. We follow the strategy outlined in [51] which is based on a nonlocal adaptation of De Giorgi’s iteration technique (see [22, 23]) but present the proof in all details due to our special treatment of the tail term.
Lemma 2.4
(truncated \(L^{\infty }-L^2\)-estimate) Assume that k satisfies \(k_{\le }\), \(\mathcal {E}_{\ge }\). There exists a constant \(C > 0\) depending on \(d,\alpha ,\lambda ,\Lambda \) such that for every \(t_0 \in (0,T)\), \(x_0 \in \mathbb {R}^d\), and \(\rho , R > 0\) with \(R \le \rho /2 \wedge t_0^{1/\alpha }\), and every subsolution u to \(\partial _t u - L_t^{\rho } u = 0\) in \(I_R(t_0) \times B_R(x_0)\) it holds:
For the definition of a subsolution to \(\partial _t u - L_t^{\rho } u = 0\) in \(I_R(t_0) \times B_R(x_0)\), we refer to the appendix.
Proof
To keep notation simple, from now on we will write \(I_R {:}{=} I_R(t_0)\), and \(B_R(x_0) : = B_R\). We do not provide the details for \(d = 1\). From the usual Caccioppoli inequality, see [51], we know that for every r, R with \(0 < r \le R \le \rho /2\) and every \(l > 0\), it holds
where \(c_1 > 0\) is a constant, \(w_l = (u-l)_+\), and \(\sigma (r,R) = r^{-(\alpha \vee 1)}(R+r)^{(\alpha \vee 1) - \alpha }\). Define \(\kappa = 1 + \frac{\alpha }{d}\), and \(A(l,R) : = \vert \lbrace (t,x) \in I_{R} \times B_R : u(t,x) > l \rbrace \vert \). Then, by \(\mathcal {E}_{\ge }\) and the fractional Sobolev inequality:
for some \(c_2,c_3,c_4 > 0\). Observe that for some constant \(c_5 > 0\) by \(k_{\le }\)
Let us now fix \(R \in (0,\rho /2]\) and define sequences \(l_i = M(1-2^{-i})\), for \(M > 0\) to be defined later, \(r_i = 2^{-i-1}R\), \(R_{i+1} = R_i - r_{i+1}\), \(R_0 {:}{=} R\), \(A_i = \int _{I_{R_i}} \int _{B_{R_i}} w_{l_i}^2(t,x) \text {d}x \text {d}t\). Note that by definition: \(R/2 = \lim _{i \rightarrow \infty } R_i< \dots< R_2< R_1 < R_0 = R\), and \(l_i \nearrow M\), and \(\sigma (r_i,R_i) \le c_6 R^{-\alpha }2^{2i}\). Then, we deduce from the two lines above:
where \(c_7,c_8,c_9 > 0\), and \(\gamma = 2 + \frac{2}{\kappa '} + d + \alpha > 0\). Let us now choose
where \(c_{10} = 2c_9\), \(C = 2^{\gamma } > 1\). Consequently, it holds
By Lemma 7.1 in [31], we obtain
for some \(c_{11}, c_{12} > 0\), as desired, where we used \(R \le \rho /2\). \(\square \)
3 Nonlocal Aronson method
In this section we prove Theorem 1.1. As is standard for proofs of heat kernel bounds for nonlocal operators, we first establish bounds for the heat kernel corresponding to the truncated jumping kernel and derive the estimate for the original jumping kernel by gluing together short and long jumps with the help of Lemma 2.2 in a second step.
The following lemma is a nonlocal version of (1.5). Presumably, estimates of this form are also of interest in the study of fractional porous medium equations and gradient flows.
Lemma 3.1
Assume that k satisfies \(k_{\le }\) and let \(\rho > 0\), \(0 \le \eta< s < T\). Let \(u \in L^{\infty }((\eta ,T) \times \mathbb {R}^d)\) be a solution to the \(\rho \)-truncated Cauchy problem (2.5) in \((\eta ,T) \times \mathbb {R}^d\). Then there exists a constant \(C = C(\Lambda )\) such that for every bounded function \(H : [\eta ,s] \times \mathbb {R}^d \rightarrow [0,\infty )\) satisfying
-
\(C\Gamma ^{\alpha }_{\rho }(H^{1/2},H^{1/2}) \le -\partial _t H\) in \((\eta ,s) \times \mathbb {R}^d\),
-
\(H^{1/2} \in L^2((\eta ,s);H^{\alpha /2}(\mathbb {R}^d))\),
the following estimate holds true:
Proof
Let \(R \ge 2\) and \(\gamma _R \in C^{\infty }_c(\mathbb {R}^d)\) such that \(\gamma _R \equiv 1\) in \(B_{R-1}(0)\), \(\gamma _R \equiv 0\) in \(\mathbb {R}^d \setminus B_{R}(0)\), \(0 \le \gamma _R \le 1\), \(\vert \nabla \gamma _R\vert \le 2\). Consequently, \(\Gamma _{\rho }^{\alpha }(\gamma _R,\gamma _R)\) satisfies:
for some constant \(c_1 > 0\). We test the equation for u with the test function \(\phi = \gamma _R^2Hu\) and integrate in time over \((\eta ,\tau )\), where \(\tau \in (\eta ,s)\), and obtain:
Note that \(\phi \) is a valid test function since for every \(t \in (\eta ,s)\) by assumption it holds \(\gamma _R H(t)u(t) \in H^{\alpha /2}(\mathbb {R}^d)\). From \((\partial _t u)u = \frac{1}{2}\partial _t(u^2)\) and integration by parts:
We treat the nonlocal term by making use of the following algebraic inequality:
where \(c_2 > 0\). Its proof is based on the following two observations:
Moreover, note that for \(R > 4\rho \) it holds \(\gamma _R^2(z) = \gamma _R^2(x) = 1\) for every \(x \in B_{\frac{R-1}{2}}(0)\) and \(z \in B_{\rho }(x)\). By symmetry of k, \(k_{\le }\) and the observations from above, we deduce:
Now, assume that H satisfies the assumption with \(C = 4c_2\Lambda \). Then, using also (3.2), the following holds true:
Upon the observation that \(\gamma _R \rightarrow 1\), as \(R \rightarrow \infty \), it follows
as desired. \(\square \)
Our next goal is to establish the following auxiliary estimate:
Theorem 3.2
Assume that k satisfies \(k_{\le }\), \(\mathcal {E}_{\ge }\). Let \(y \in \mathbb {R}^d\), \(\sigma ,\rho > 0\), \(\eta \ge 0\). Let \(u_0 \in L^2(\mathbb {R}^d)\) be such that \(u_0 \equiv 0\) in \(B_{\sigma }(y)\). Assume that \(u \in L^{\infty }( (\eta ,T) \times \mathbb {R}^d)\) is a weak solution to \(\partial _t u - L_t^{\rho } u = 0\) in \((\eta ,T) \times \mathbb {R}^d\). Then there exist \(\nu> 1, C > 0\) depending on \(d,\alpha ,\lambda ,\Lambda \) such that for every \(s \in (\eta ,T)\) with \( s-\eta \le \frac{1}{4\nu }\rho ^{\alpha }\):
The idea to prove Theorem 3.2 is to find a suitable function H such that Lemma 3.1 is applicable. In the following, we present a suitable such function.
Given \(y \in \mathbb {R}^d\), \(\rho > 0\), \(0 \le \eta< s < T\), \(\nu > 1\) with \(s-\eta \le \frac{1}{4\nu }\rho ^{\alpha }\), we define \(H_{y,\rho ,\eta ,s,\nu } = H : [\eta ,s] \times \mathbb {R}^d \rightarrow [0,\infty )\) via
Lemma 3.3
For every \(C > 0\), there exists \(\nu = \nu (d,\alpha , C) > 1\) such that for every \(y \in \mathbb {R}^d\), \(\rho > 0\), \(0 \le \eta< s < T\) with \(s-\eta \le \frac{1}{4\nu }\rho ^{\alpha }\), the function \(H_{y,\rho ,\eta ,s,\nu } = H\) defined above satisfies
Proof
Let \(y \in \mathbb {R}^d\), \(\rho > 0\), \(0 \le \eta< s <T\), with \(s-\eta \le \frac{1}{4\nu }\rho ^{\alpha }\), where \(\nu > 1\) to be chosen later. Note that by assumption, \(\frac{\rho ^{\alpha }}{\nu [2(s-\eta ) - (t-\eta )]} > 1\) for every \(t \in [\eta ,s]\). Let \(t \in [\eta ,s]\) be fixed. We split the proof of (3.4) into three cases.
Case 1: \(\vert x-y \vert \le 2 \rho \).
In this case, trivially \(\Gamma ^{\alpha }_{\rho }(H^{1/2},H^{1/2})(t,x) = 0\), and
Therefore, (3.4) holds true for any \(\nu > 0\).
Case 2: \(2\rho \le \vert x-y \vert \le 3\rho \).
In this case, \(-\partial _t H(t,x) = \nu \rho ^{-\alpha }\), as in Case 1. Moreover, let us fix \(x_0 \in B_{\rho }(x)\) with \(\vert x_0 - y\vert = 3\rho \) such that for every \(z \in \mathbb {R}^d \setminus B_{3\rho }(y)\) it holds \(\vert x_0-z \vert \le 2 \vert x-z \vert \). This holds true, e.g. if one chooses \(x_0\) as a point on \(\partial B_{3\rho }(y)\) that minimizes \({{\,\textrm{dist}\,}}(x,\mathbb {R}^d \setminus B_{3\rho }(y))\), since then by triangle inequality: \(\vert x_0 - z\vert \le \vert x_0 - x\vert + \vert x - z\vert \le 2 \vert x-z \vert \). Note that \(H(t,x) = H(t,x_0)\) and \(B_{\rho }(x) \subset B_{2\rho }(x_0)\).
Therefore:
for some \(c_1,c_2,c_3,c_4 > 0\). We used the inequality \(\log (a) \le a^{1/2}\) in the last step. Moreover, note that the estimate \(\left( H^{1/2}(t,x_0) - H^{1/2}(t,z) \right) ^2 \le \vert \nabla H^{1/2}(t,x_0) \vert ^2 \vert x_0 - z \vert ^{2}\) is correct since \(\sup _{z \in \mathbb {R}^d \setminus B_{3\rho }(y)} \vert \nabla H^{1/2}(t,z)\vert = \vert \nabla H^{1/2}(t,x_0)\vert \) due to \(x_0 \in \partial B_{3\rho }(y)\). Therefore, (3.4) holds true in this case for any \(\nu > c_4C\).
Case 3: \(\vert x-y \vert > 3\rho \).
In this case,
Moreover:
for some \(c_5,c_6,c_7 > 0\). In the third inequality, we used \(\vert z-y \vert \ge \vert x-y \vert - \rho \), and in the second to last step we applied the estimate \(\log (a) \le c a^{1/3}\). This holds with \(c > 0\) independent of \(a > 1\). Therefore, by choosing \(\nu > c_7C\), (3.4) is satisfied also in this case.
Together, we have proved (3.4). Finally, note that \(\Gamma ^{\alpha }_{\rho }(H^{1/2},H^{1/2}) \in L^1((\eta ,s) \times \mathbb {R}^d)\) since for \(\vert x-y \vert \ge 3\rho \), we computed above
where \(c > 1\) is a constant that might depend on \(\eta ,s,\rho \). This proves (3.5). \(\square \)
Having at hand the function H defined in (3.3), it is possible to establish Theorem 3.2.
Proof
(Proof of Theorem 3.2) The idea is to apply Lemma 3.1 with H as in Lemma 3.3. It follows that for every \(y \in \mathbb {R}^d\), \(0 \le \eta< s < T\) with \(s-\eta \le \frac{1}{4\nu }\rho ^{\alpha }\):
Consequently,
By the truncated \(L^{\infty }-L^2\)-estimate Lemma 2.4, applied with \(R = (s-\eta )^{1/\alpha }\), \(t_0 = s\), \(x_0 = y\):
for some \(c_1 > 0\), where \(\eta ' {:}{=} s - 2^{-\alpha }(s-\eta ) \in (\eta ,s)\).
Note that there exist \(c_2,c_3 > 0\) such that for \(x \in \mathbb {R}^d \setminus B_{\sigma }(y)\) it holds
and for \((\tau ,x) \in [\eta ,s] \times B_{2\rho }(y)\) we have:
This follows directly from the definition of H and \(s-\eta \le \frac{1}{4\nu }\rho ^{\alpha }\). Together, we obtain
for some \(c_4 > 0\), as desired. \(\square \)
Having proved Theorem 3.2, we are now in the position to establish upper off-diagonal bounds for \(p_{\rho }(y,s;x,\eta )\):
Theorem 3.4
Assume that k satisfies \(k_{\le }\), \(\mathcal {E}_{\ge }\). Then there exists \(c > 0\) depending on \(d,\alpha ,\lambda ,\Lambda \) such that for every \(\rho > 0\), \(0 \le \eta< s < T\), and \(x,y \in \mathbb {R}^d\) with \(s-\eta \le \frac{1}{4\nu }\rho ^{\alpha }\):
Proof
Note that the on-diagonal bound (2.14) and (2.7) immediately imply for every \(0 \le \eta< s < T\) with \(s-\eta \le \frac{1}{4\nu }\rho ^{\alpha }\) and \(x,y \in \mathbb {R}^d\):
for some \(c_1 > 0\). On the other hand, from Theorem 3.2, it follows for every \(0 \le \eta< s < T\) with \(s-\eta \le \frac{1}{4\nu }\rho ^{\alpha }\) and \(x,y \in \mathbb {R}^d\):
for some \(c_2 > 0\). To see this, one observes that \(u(t,x) = \int _{\mathbb {R}^d \setminus B_{\sigma }(y)} p_{\rho }(x,t;z,\eta )p_{\rho }(y,s;z,\eta )\text {d}z\) satisfies the assumptions of Theorem 3.2 with
To prove (3.6), let us fix \(0 \le \eta< s < T\) with \(s-\eta \le \frac{1}{4\nu }\rho ^{\alpha }\), and \(x,y \in \mathbb {R}^d\). Then we define \(\sigma = \frac{1}{2}\vert x-y \vert \) and compute, using (2.8):
For \(J_1\), we compute, using (3.7), (3.8):
for some \(c_3 > 0\). For \(J_2\), observe that \(B_{\sigma }(y) \subset \mathbb {R}^d \setminus B_{\sigma }(x)\), and therefore by (2.6):
for some \(c_4 > 0\). Together, we obtain the desired result. \(\square \)
Bounds for the heat kernel corresponding to the truncated jumping kernel \(p_{\rho }\) imply bounds for p with the help of the gluing lemma Lemma 2.2. The underlying argument is known among probabilists as “Meyer’s decomposition”. For an analytic proof relying on the parabolic maximum principle we refer to the appendix.
We are now ready to provide the proof of our main result Theorem 1.1:
Proof of Theorem 1.1
Let \(x,y \in \mathbb {R}^d\) be fixed. By (2.13) it suffices to prove that for some constants \(c_0, c_1 > 0\) and \(s-\eta \le c_0 \vert x-y \vert ^{\alpha }\) it holds
By Lemma 2.2, and \(k_{\le }\) we know that for every \(\rho > 0\), \(0 \le \eta< s < T\):
for some \(c_2,c_3 > 0\). We choose \(\rho = \frac{\vert x-y \vert }{12}\left( \frac{d+\alpha }{\alpha } + \frac{1}{2} + \frac{d}{2\alpha }\right) ^{-1}\). Then by (3.6) and (3.9) it holds for \(s-\eta \le \frac{1}{4\nu }\rho ^{\alpha }\):
where \(c_4 > 0\), as desired. \(\square \)
4 Extension: jumping kernels of mixed type on metric measure spaces
In this section, we discuss a possible extension of the nonlocal Aronson method to jumping kernels of mixed type. Moreover, we work on a general doubling metric measure space. The main result of this section is Theorem 4.1.
Let (M, d) be a locally compact, separable metric space, and let \(\mu \) be a positive Radon measure with full support. We assume that \((M,d,\mu )\) satisfies the volume doubling property, i.e., there exists \(C > 0\), \(d \in \mathbb {N}\) such that
Note that as a consequence of VD, for every \(\delta > 0\) there exist \(c_1,c_2 > 0\) such that for every \(R > 0\), \(x,y \in M\) with \(d(x,y) \le \delta R\): \(c_1 \mu (B_{R}(x)) \le \mu (B_{R}(y)) \le c_2 \mu (B_{R}(x))\).
Moreover, let \(\phi : [0,\infty ) \rightarrow [0,\infty )\) be strictly increasing with \(\phi (0) = 0\), \(\phi (1) = 1\) and
for some constant \(C > 0\) and \(0< \alpha _1 \le \alpha _2 < 2\).
For a detailed discussion of the setup, we refer to [17].
Consider symmetric jumping kernels \(k : (0,T) \times M \times M \rightarrow \mathbb {R}\) satisfying for some \(\Lambda > 0\)
and assume that there is \(\mathcal {F} \subset L^2(M,\mu )\) such that \((\mathcal {E}_t,\mathcal {F})\) is a regular Dirichlet form on \(L^2(M,\mu )\) for every \(t \in (0,T)\), where for every \(u,v \in \mathcal {F}\),
is defined in the usual way. For simplicity, we write \(\text {d}x {:}{=} \mu (\text {d}x)\).
Moreover, we assume that the Faber-Krahn inequality holds true, i.e., that there exist \(c, \nu > 0\) such that for all \(t \in (0,T)\), \(R > 0\), \(x_0 \in M\), \(D \subset B_R(x_0)\) and every \(u \in \mathcal {F}\) with \(u \equiv 0\) in \(M \setminus D\):
Let \(L_t\) be the operator associated with \(\mathcal {E}_t\) and \(p(y,s;x,\eta )\) be the fundamental solution to the equation \(\partial _t u - L_t u = 0\).
Theorem 4.1
Let \((M,d,\mu )\) and \(\phi \) be as above, and assume VD, (4.1). Assume that k satisfies \(k_{\le }\) and FK. Then there exists \(c > 0\) such that for every \(0 \le \eta< s < T\), \(x,y \in M\):
We remark that variants of Theorem 4.1 for time-homogeneous jumping kernels of mixed type on doubling metric measure spaces can be found in several articles, e.g., [14, 15, 17].
We will provide a proof of Theorem 4.1 below. A central ingredient in the proof is the \(L^{\infty }-L^2\)-estimate (4.8) for subsolutions to \(\partial _t u- L_t^{\rho } u = 0\). Its proof is similar to the proof of Lemma 2.4. However we will provide some details since this seems to be the first time that FK is used for nonlocal parabolic \(L^{\infty }-L^2\)-estimates. Moreover, we provide an \(L^{\infty }-L^1\)-estimate for subsolutions to \(\partial _t u - L_t u = 0\) which allows us to give a direct proof of the on-diagonal upper heat kernel estimate. In the elliptic case, \(L^{\infty }-L^2\)- and \(L^{\infty }-L^1\)-estimates are established via FK for example in [17].
Lemma 4.2
Let \((M,d,\mu )\) and k be as in Theorem 4.1. Then there exists \(C_1 > 0\) such that for every \(t_0 \in (0,T)\), \(x_0 \in M\), \(\rho , R > 0\) with \(R \le \rho /2 \wedge \phi ^{-1}(t_0)\) and every subsolution u to \(\partial _t u - L_t^{\rho } u = 0\) in \(I_R(t_0) \times B_R(x_0)\) it holds:
Moreover, there exists \(C_2 > 0\) such that for every \(t_0 \in (0,T)\), \(x_0 \in M\), \(R \le \phi ^{-1}(t_0)\) and every subsolution u to \(\partial _t u - L_t u = 0\) in \(I_R(t_0) \times B_R(x_0)\), with \(u \ge 0\) in \(I_R(t_0) \times B_R(x_0)\), it holds:
We refer to Sect. 1 for the definition of a subsolution. \(H^{\alpha /2}(\mathbb {R}^d)\) should be replaced by \(\mathcal {F}\).
Proof
First, we prove (4.3). Let \(l> k > 0\), \(0 < r \le R \le \rho /2\), \(A(l,R) : = \vert \lbrace (t,x) \in I_{R} \times B_R(x_0) : u(t,x) > l \rbrace \vert \). Let u be a subsolution to \(\partial _t u - L_t^{\rho } u = 0\). First, observe that for every \(t \in I_{R+r}\):
where \(c_1 > 0\), \(\sigma '(R,r) {:}{=} \sup _{x \in B_R(x_0)} \phi (r)^{-1} \frac{\mu (B_{\rho }(x_0))}{\mu (B_{r}(x))}\) and we used \(k_{\le }\). By Caccioppoli’s inequality and (4.7), every subsolution u to \(\partial _t u - L_t^{\rho } u = 0\) in \(I_R \times B_R(x_0)\) satisfies:
where \(c_2 > 0\), \(\sigma (R,r) = \phi (r)^{-1} \vee (\phi (R+r)-\phi (R))^{-1}\) and
where \(c_3,c_4 > 0\), \(\tau \in C^{\infty }_c(\mathbb {R}^d)\) is an arbitrary function such that \(\tau \equiv 1\) in \(B_R(x_0)\), \(\tau \equiv 0\) in \(B_{R+\frac{r}{2}}(x_0)\), \(\Vert \nabla \tau \Vert _{\infty } \le 4 r^{-1}\), and we used FK. By combination of the foregoing two estimates, we obtain for some \(c_5 > 0\):
Let us now fix \(R \in (0,\rho /2]\) and define sequences \(l_i = M(1-2^{-i})\), for \(M > 0\) to be defined later, \(r_i = 2^{-i-1}R\), \(R_{i+1} = R_i - r_{i+1}\), \(R_0 {:}{=} R\), \(A_i = \int _{I_{R_i}} \int _{B_{R_i}(x_0)} w_{l_i}^2(t,x) \text {d}x \text {d}t\). We deduce:
for some \(\gamma ,c_6,c_7 > 0\), using that \(\sigma (R_i,r_i) \le c_8 2^{\gamma _1 i} \phi (R)^{-1}\), \(\sigma '(R_i,r_i) \le c_9 2^{\gamma _2 i} \phi (R)^{-1} \frac{\mu ( B_{\rho }(x_0))}{\mu ( B_{R}(x_0))}\) for some \(c_8,c_9,\gamma _1,\gamma _2 > 0\). The latter follows from the fact that for all \(x \in B_{R}(x_0)\): \(\frac{\mu (B_{\rho }(x_0))}{\mu (B_{r_i}(x))} = \frac{\mu (B_{\rho }(x_0))}{\mu (B_{R}(x))} \frac{\mu (B_{R}(x))}{\mu (B_{r_i}(x))} \le c_{10} 2^{\gamma _3 i}\frac{\mu (B_{\rho }(x_0))}{\mu (B_{R}(x_0))} \), where \(c_{10},\gamma _3 > 0\). Let us choose \(c_{11} = c_7 2^{1+\gamma }\), and
Hence:
and we can apply Lemma 7.1 in [31] to deduce that for some \(c_{12} > 0\):
This proves (4.3). Let us now demonstrate how to prove (4.4). Let u be a subsolution to \(\partial _t u - L_t u = 0\). First, we provide a different estimate of the tail term. For every \(t \in I_{R+r}\):
where \({\widetilde{\sigma }}'(R,r) {:}{=} \sup _{x \in B_{R}(x_0)} \mu (B_r(x))^{-1} \phi (r)^{-1}\) and we applied \(k_{\le }\). As in (4.5), we get:
for some \(c_{14} > 0\). From now on, let \({\overline{R}} > 0\) be fixed. Moreover, let \(0< {\overline{R}}/2 \le r < R \le {\overline{R}}\) and define sequences \(l_i = M(1-2^{-i})\), for \(M > 0\) to be defined later, \(r_i = 2^{-i-1}(R-r)\), \(R_{i+1} = R_i - r_{i+1}\), \(R_0 {:}{=} R\), \(A_i = \int _{I_{R_i}} \int _{B_{R_i}(x_0)} w_{l_i}^2(t,x) \text {d}x \text {d}t\) and deduce
for \(c_{16},c_{17},\gamma > 0\), using (4.1) and that by VD: \(\mu (B_{R_i}(x_0)) \ge \mu (B_{{\overline{R}}/2}(x_0)) \ge c_{18} \mu (B_{{\overline{R}}}(x_0)) \ge c_{18}\mu (B_{R}(x_0))\), \(\sigma (R_i,r_i) \le c_{19}2^{\gamma _4 i}\frac{R}{R-r}\phi (R-r)^{-1}\), \({\widetilde{\sigma }}'(R_i,r_i) \le c_{20} 2^{\gamma _5 i} \phi (R-r)^{-1} \left( \frac{R}{R-r} \right) ^d \mu (B_R(x_0))^{-1}\). The latter follows from the fact that for all \(x \in B_R(x_0)\): \(\mu (B_{r_i}(x))^{-1} = \frac{\mu (B_R)(x)}{\mu (B_{r_i}(x))} \mu (B_R(x))^{-1} \le c_{21} 2^{\gamma _6 i} \left( \frac{R}{R-r} \right) ^d \mu (B_R(x_0))^{-1}\). We choose \(c_{22} = c_{17}2^{1+\gamma }\) and
and deduce by arguments analogous to those in the first part of the proof:
where \(c_{23} > 0\). We further estimate
where \(c_{24} > 0\) and we applied (4.1) and Hölder’s and Young’s inequality. Together, we obtain
for \(c_{25} > 0\) and \(\delta {:}{=} d-1 \vee \frac{\alpha _2(1 + \nu ) + 1}{\nu }\). We can apply Lemma 1.1 in [26] to the estimate above and deduce that there exists \(c_{26} > 0\) such that for every \(0< {\overline{R}}/2 \le r < R \le {\overline{R}}\):
Choosing \(r = {\overline{R}}/2, R = {\overline{R}}\) implies the desired result (4.4). \(\square \)
Now, having established the \(L^{\infty }-L^2\)-estimate Lemma 4.2, we are ready to prove Theorem 4.1.
Proof of Theorem 4.1
First of all, we observe that by VD, (4.1), \(k_{\le }\):
for every \(t \in (0,T)\), \(x \in M\), \(R > 0\). For a proof, see [17]. Given \(y \in M\), \(\rho > 0\), \(0 \le \eta< s < T\), \(\nu > 1\) with \(s-\eta \le \frac{1}{4\nu }\phi (\rho )\), we define \(H_{y,\rho ,\eta ,s,\nu } = H : [\eta ,s] \times M \rightarrow [0,\infty )\) via
With the help of (4.7), (4.1) it is easy to check along the lines of Lemma 3.3 that H satisfies the assumptions of Lemma 3.1, namely for every \(C > 0\) there exists \(\nu > 1\) such that
where
Next, we recall from Lemma 4.2 that the following \(L^{\infty }-L^2\)-estimate holds true for local subsolutions u to \(\partial _t u - L_t^{\rho } u = 0\) in \(I_R(t_0) \times B_R(x_0)\), where \(I_R(t_0) = (t_0 - \phi (R),t_0)\):
Applying H to Lemma 3.1, we obtain from (4.8) and the definition of H:
for \(c_1 > 0\), where \(u,u_0\) are as in Lemma 3.1, and \(0 \le \eta< s < T\) with \(s- \eta \le \frac{1}{4\nu } \phi (\rho )\), \(\sigma > 0\).
Moreover, the following on-diagonal estimates hold for every \(0 \le \eta< s < T\), \(x,y \in M\), \(\rho > 0\):
for some \(c > 0\). These estimates are proved in Lemma 5.1 and Section 4.4 in [17] using a stochastic approach. A more direct proof, using only analysis tools, goes via the \(L^{\infty }-L^1\)-estimate (4.4).
In fact, given \(0 \le \eta< s < T\), \(x,y \in M\), we apply (4.4) to \((t,z) \mapsto p(y,t;z,\eta )\), choosing \(R {:}{=} \phi ^{-1}(s-\eta )\), \(x_0 {:}{=} x\), \(t_0 {:}{=} s\). Using (2.7), we obtain
as desired. (4.11) is a direct consequence of (4.10) in the light of (5.7) and (4.7). Note that the proof of (5.7) is written for \(M = \mathbb {R}^d\) but works in the same way in the current setup.
Combining (4.9) and (4.11), we derive as in Theorem 3.4 for every \(0 \le \eta< s < T\) with \(s- \eta \le \frac{1}{4\nu } \phi (\rho )\), \(x,y \in M\):
for \(c_2 > 0\). Finally, we explain how to deduce off-diagonal bounds for p. We get from (5.6):
We choose \(\rho = \frac{d(x,y)}{12}\left( \frac{d+\alpha _1}{\alpha _1} + \frac{1}{2} + \frac{d}{2\alpha _1}\right) ^{-1}\) and obtain by VD, (4.1):
for \(c_3,c_4,c_5 > 0\). Next, we estimate \(\int _{\eta }^s P_{\tau }^{\rho } K_{\rho }(y) \text {d}\tau \). For this, we compute by \(k_{\le }\) and VD:
for \(c_6,c_7 > 0\), and \(c > 3 + \frac{6d}{\alpha _1}\). Using (4.9), VD and (4.1), we estimate for \(\tau \in (\eta ,s)\), \(k \ge 2\):
for \(c_8,c_9,c_{10} > 0\), where we use \(s- \eta \le \frac{1}{4\nu } \phi (\rho )\). From (2.7), it follows
for \(c_{11},c_{12} > 0\). Combining (4.12), (4.13), (4.14) we obtain the desired off-diagonal estimate for \(s-\eta \le \frac{1}{4\nu }\phi (\rho )\). Together with the on-diagonal estimate (4.10), we deduce the desired result. \(\square \)
Remark
Note that the proof of Theorem 4.1 does not require the scaling argument from [14] since we are working with an on-diagonal estimate and an \(L^{\infty }-L^2\)-estimate that take into account the parabolic scaling of the corresponding equation, see also [15].
Data availability
Data sharing is not applicable to this article as no data sets were generated or analysed.
References
Aronson, D.G.: Bounds for the fundamental solution of a parabolic equation. Bull. Am. Math. Soc. 73, 890–896 (1967)
Aronson, D.G.: Non-negative solutions of linear parabolic equations. Ann. Scuola Norm. Sup. Pisa Cl. Sci. (3) 22, 607–694 (1968)
Aronson, D.G., Serrin, J.: Local behavior of solutions of quasilinear parabolic equations. Arch. Ration. Mech. Anal. 25, 81–122 (1967)
Aronson, D.G., Bénilan, P.: Régularité des solutions de l’équation des milieux poreux dans \(\mathbb{R}^n\). C. R. Acad. Sci. Paris Sér. A-B 288, A103–A105 (1979)
Barlow, M.T., Bass, R.F., Kumagai, T.: Parabolic Harnack inequality and heat kernel estimates for random walks with long range jumps. Math. Z. 261(2), 297–320 (2009)
Bonforte, M., Dolbeault, J., Nazaret, B., Simonov, N.: Stability in Gagliardo–Nirenberg–Sobolev inequalities: flows, regularity and the entropy method (2020). arXiv:2007.03674
Bonforte, M., Endal, J.: Nonlocal nonlinear diffusion equations. Smoothing effects, Green functions, and functional inequalities. J. Funct. Anal. (2022). arXiv:2205.06850
Blumenthal, R.M., Getoor, R.K.: Some theorems on stable processes. Trans. Am. Math. Soc. 95, 263–273 (1960)
Barlow, M.T., Grigor’yan, A., Kumagai, T.: Heat kernel upper bounds for jump processes and the first exit time. J. Reine Angew. Math. 626, 135–157 (2009)
Bae, J., Kang, J., Kim, P., Lee, J.: Heat kernel estimates for symmetric jump processes with mixed polynomial growths. Ann. Probab. 47(5), 2830–2868 (2019)
Bass, R.F., Levin, D.A.: Transition probabilities for symmetric jump processes. Trans. Am. Math. Soc. 354(7), 2933–2953 (2002)
Bonforte, M., Sire, Y., Vázquez, J.L.: Optimal existence and uniqueness theory for the fractional heat equation. Nonlinear Anal. 153, 142–168 (2017)
Chen, Z.-Q., Kumagai, T.: Heat kernel estimates for stable-like processes on \(d\)-sets. Stoch. Process. Appl. 108(1), 27–62 (2003)
Chen, Z.-Q., Kumagai, T.: Heat kernel estimates for jump processes of mixed types on metric measure spaces. Probab. Theory Relat. Fields 140(1–2), 277–317 (2008)
Chen, Z.-Q., Kim, P., Kumagai, T., Wang, J.: Heat kernel upper bounds for symmetric Markov semigroups. J. Funct. Anal. 281(4):Paper No. 109074, 40 (2021)
Carlen, E.A., Kusuoka, S., Stroock, D.W.: Upper bounds for symmetric Markov transition functions. Ann. Inst. H. Poincaré Probab. Stat. 23(2, suppl.), 245–287 (1987)
Chen, Z.-Q., Kumagai, T., Wang, J.: Stability of heat kernel estimates for symmetric non-local Dirichlet forms. Mem. Am. Math. Soc. 271(1330), v+89 (2021)
Cheng, S.Y., Li, P., Yau, S.T.: On the upper estimate of the heat kernel of a complete Riemannian manifold. Am. J. Math. 103(5), 1021–1063 (1981)
Chaker, J., Silvestre, L.: Coercivity estimates for integro-differential operators. Calc. Var. Partial Differ. Equ. 59(4):Paper No. 106, 20 (2020)
Davies, E.B.: Heat Kernels and Spectral Theory, Volume 92 of Cambridge Tracts in Mathematics. Cambridge University Press, Cambridge (1989)
Dyda, B., Kassmann, M.: Regularity estimates for elliptic nonlocal operators. Anal. PDE 13(2), 317–370 (2020)
Di Castro, A., Kuusi, T., Palatucci, G.: Nonlocal Harnack inequalities. J. Funct. Anal. 267(6), 1807–1836 (2014)
Di Castro, A., Kuusi, T., Palatucci, G.: Local behavior of fractional \(p\)-minimizers. Ann. Inst. H. Poincaré Anal. Non Linéaire 33(5), 1279–1299 (2016)
de Pablo, A., Quirós, F., Rodriguez, A., Vázquez, J.L.: A general fractional porous medium equation. Commun. Pure Appl. Math. 65(9), 1242–1284 (2012)
Felsinger, M., Kassmann, M., Voigt, P.: The Dirichlet problem for nonlocal operators. Math. Z. 279(3–4), 779–809 (2015)
Giaquinta, M., Giusti, E.: On the regularity of the minima of variational integrals. Acta Math. 148, 31–46 (1982)
Grigor’yan, A., Jiaxin, H.: Off-diagonal upper estimates for the heat kernel of the Dirichlet forms on metric spaces. Invent. Math. 174(1), 81–126 (2008)
Grigor’yan, A., Eryan, H., Jiaxin, H.: Lower estimates of heat kernels for non-local Dirichlet forms on metric measure spaces. J. Funct. Anal. 272(8), 3311–3346 (2017)
Grigor’yan, A., Eryan, H., Jiaxin, H.: Two-sided estimates of heat kernels of jump type Dirichlet forms. Adv. Math. 330, 433–515 (2018)
Grigor’yan, A., Jiaxin, H., Lau, K.-S.: Estimates of heat kernels for non-local regular Dirichlet forms. Trans. Am. Math. Soc. 366(12), 6397–6441 (2014)
Giusti, E.: Direct Methods in the Calculus of Variations. World Scientific Publishing Co., Inc., River Edge (2003)
Grigor’yan, Alexander: Heat Kernel and Analysis on Manifolds, Volume 47 of AMS/IP Studies in Advanced Mathematics. American Mathematical Society, Providence (2009)
Jiaxin, H., Li, X.: The Davies method revisited for heat kernel upper bounds of regular Dirichlet forms on metric measure spaces. Forum Math. 30(5), 1129–1155 (2018)
Kang, J.: Heat kernel estimates for symmetric jump processes with anisotropic jumping kernels. Proc. AMS 151(01), 385–399 (2023)
Kigami, J.: Local Nash inequality and inhomogeneity of heat kernels. Proc. Lond. Math. Soc. (3) 89(2), 525–544 (2004)
Kassmann, M., Kim, K.-Y., Kumagai, T.: Heat kernel bounds for nonlocal operators with singular kernels. J. Math. Pures Appl. 9(164), 1–26 (2022)
Komatsu, T.: Continuity estimates for solutions of parabolic equations associated with jump type Dirichlet forms. Osaka J. Math. 25(3), 697–728 (1988)
Komatsu, T.: Uniform estimates for fundamental solutions associated with non-local Dirichlet forms. Osaka J. Math. 32(4), 833–860 (1995)
Li, P.: Geometric Analysis. Cambridge Studies in Advanced Mathematics, vol. 134. Cambridge University Press, Cambridge (2012)
Lu, P., Ni, L., Vázquez, J.L., Villani, C.: Local Aronson-Benilan estimates and entropy formulae for porous medium and fast diffusion equations on manifolds. J. Math. Pures Appl. 91(9), 1–19 (2009)
Li, P., Yau, S.-T.: On the parabolic kernel of the Schrödinger operator. Acta Math. 156(3–4), 153–201 (1986)
Meyer, P.A.: Renaissance, recollements, mélanges, ralentissement de processus de Markov. Ann. Inst. Fourier (Grenoble), 25(3–4):xxiii, 465–497 (1975)
Maekawa, Y., Miura, H.: On fundamental solutions for non-local parabolic equations with divergence free drift. Adv. Math. 247, 123–191 (2013)
Maekawa, Y., Miura, H.: Upper bounds for fundamental solutions to non-local diffusion equations with divergence free drift. J. Funct. Anal. 264(10), 2245–2268 (2013)
Moser, J.: A Harnack inequality for parabolic differential equations. Commun. Pure Appl. Math. 17, 101–134 (1964)
Moser, J.: Correction to: “A Harnack inequality for parabolic differential equations’’. Commun. Pure Appl. Math. 20, 231–236 (1967)
Moser, J.: On a pointwise estimate for parabolic differential equations. Commun. Pure Appl. Math. 24, 727–740 (1971)
Murugan, M., Saloff-Coste, L.: Heat kernel estimates for anomalous heavy-tailed random walks. Ann. Inst. Henri Poincaré Probab. Stat. 55(2), 697–719 (2019)
Nash, J.: Continuity of solutions of parabolic and elliptic equations. Am. J. Math. 80, 931–954 (1958)
Porper, F.O., Eidel’ man, S.D.: Two-sided estimates of the fundamental solutions of second-order parabolic equations and some applications of them. Uspekhi Mat. Nauk 39(3(237)), 107–156 (1984)
Strömqvist, M.: Harnack’s inequality for parabolic nonlocal equations. Ann. Inst. H. Poincaré Anal. Non Linéaire 36(6), 1709–1745 (2019)
Vázquez, J.L.: Asymptotic behaviour for the fractional heat equation in the Euclidean space. Complex Var. Elliptic Equ. 63(7–8), 1216–1231 (2018)
Vázquez, J.L., de Pablo, A., Quirós, F., Rodríguez, A.: Classical solutions and higher regularity for nonlinear fractional diffusion equations. J. Eur. Math. Soc. (JEMS) 19(7), 1949–1975 (2017)
Zeidler, E.: Nonlinear Functional Analysis and Its Applications. II, A. Springer, New York. Linear monotone operators. Translated from the German by the author and Leo F, Boron (1990)
Funding
Open Access funding enabled and organized by Projekt DEAL. Moritz Kassmann and Marvin Weidner gratefully acknowledge financial support by the German Research Foundation (SFB 1283 - 317210226 resp. GRK 2235 - 282638148). Marvin Weidner has received funding from the European Research Council (ERC) under the Grant Agreement No 801867.
Author information
Authors and Affiliations
Contributions
All authors developed and discussed the results and contributed to the final manuscript.
Corresponding author
Ethics declarations
Conflict of interest
All authors state no conflict of interest.
Informed consent
Informed consent has been obtained from all individuals included in this research work.
Additional information
Communicated by Andrea Mondino.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix
Appendix
The main goal of this section is to give a proof of Lemma 2.2 via analysis methods. We mainly follow the strategy carried out in [27] modifying some of their arguments due to the time-inhomogeneity of the jumping kernel.
First, we introduce the notion of a subsolution to \(\partial _t u - L_t u = 0\) in \(I \times \Omega \) for some open interval \(I \subset (\eta ,T)\) and some open set \(\Omega \subset \mathbb {R}^d\).
Definition 5.1
Let \(\Omega \subset \mathbb {R}^d\) be open and bounded. We say that a function \(u \in L^2_{loc}(I;H^{\alpha /2}(\mathbb {R}^d))\) with \(\partial _t u \in L^1_{loc}(I ; L^2_{loc}(\Omega ))\) is a subsolution to
if for every \(\phi \in H^{\alpha /2}(\mathbb {R}^d)\) with \(\phi \equiv 0\) in \(\mathbb {R}^d \setminus \Omega \) and \(\phi \ge 0\):
In this case we say that u solves \(\partial _t u - L_t u \le 0\) in \(I \times \Omega \). u is a solution to (5.1) if (5.2) holds for any \(\phi \in H^{\alpha /2}(\mathbb {R}^d)\) with \(\phi \equiv 0\) in \(\mathbb {R}^d \setminus \Omega \). (Sub-)solutions to the corresponding \(\rho \)-truncated problem are defined accordingly, replacing k by \(k_{\rho }\).
The main ingredient in the proof of Lemma 2.2 is the following parabolic maximum principle, which is an analog of Proposition 4.11 in [27]:
Lemma 5.2
(parabolic maximum principle) Let \(\Omega \subset \mathbb {R}^d\) be open. Assume that k satisfies \(k_{\le }\), \(\mathcal {E}_{\ge }\). Assume that u solves
Then \(u \le 0\) a.e. in \((\eta ,T) \times \Omega \). The same result holds for subsolutions to \(\partial _t u - L_t^{\rho } u = 0\).
Proof
By assumptions \(k_{\le }\), \(\mathcal {E}_{\ge }\) and dominated convergence theorem, we have for \(f \in L^2(\mathbb {R}^d)\)
Thus, \((\mathcal {E}_t,H^{\alpha /2}(\mathbb {R}^d))\) is a regular Dirichlet form for every \(t \in (0,T)\). Along the lines of Lemma 4.3 in [27] one can prove
where \(\phi \in C^{\infty }(\mathbb {R})\) such that \(\phi \equiv 0\) in \((-\infty ,0]\), \(\phi > 0\) on \((0,\infty )\), and \(0 \le \phi ' \le 1\). From here, the remainder of the proof follows along the lines of Proposition 4.11 in [27]. The proof for subsolutions to \(\partial _t u - L_t^{\rho } u = 0\) is carried out via similar arguments, using that for some \(c > 0\):
\(\square \)
In the following, for \(f \in L^2(\Omega )\), we will denote by \((s,x) \mapsto P^{\Omega }_{\eta ,s}f(x)\) the solution to \(\partial _t u - L_t u = 0\) in \((\eta ,T) \times \Omega \) with \(\Vert P^{\Omega }_{\eta ,s}f - f \Vert _{L^2(\Omega )} \rightarrow 0\) as \(s \searrow \eta \), and \(P^{\Omega }_{\eta ,s}f \equiv 0\) in \(\mathbb {R}^d \setminus \Omega \). We define \((s,x) \mapsto P^{\Omega ,\rho }_{\eta ,s}f(x)\) to be solution to the corresponding \(\rho \)-truncated problem.
In order to prove Lemma 2.2, we need an approximation result for \(P_{\eta ,s}f\). In the time-homogeneous case, its proof is given in Lemma 4.17 in [27]. However, their argument does not work in our situation due to the lack of a resolvent operator associated with \(P_{\eta ,s}\).
Lemma 5.3
Let \((\Omega _n)_{n \in \mathbb {N}}\) be an increasing sequence of open subsets of \(\mathbb {R}^d\) with \(\bigcup _{n \in \mathbb {N}} \Omega _n = \mathbb {R}^d\), and \(\eta \ge 0\). Assume that k satisfies \(k_{\le }\), \(\mathcal {E}_{\ge }\). Then for every \(f \in L^2(\mathbb {R}^d)\) it holds \(P_{\eta ,s}^{\Omega _n} f \rightarrow P_{\eta ,s} f\), \(P_{\eta ,s}^{\Omega _n,\rho } f \rightarrow P^{\rho }_{\eta ,s} f\) pointwise a.e. for a.e. \(s \in (\eta ,T)\), and every \(\rho > 0\).
Proof
We denote \(u_n(s) {:}{=} P^{\Omega _n}_{\eta ,s} f\). By the parabolic maximum principle it follows that \(u_n(s) \le u_{n+1}(s) \le \cdots \le P_{\eta ,s} f\). Consequently, there exists a function \(u \in L^2((\eta ,T)\times \mathbb {R}^d)\) such that \(u_n(s) \nearrow u(s) \le P_{\eta ,s} f\) a.e., and by dominated convergence we have \(\Vert u_n - u\Vert _{L^2((\eta ,T)\times \mathbb {R}^d)} \rightarrow 0\).
It remains to prove \(u(s) = P_{\eta ,s}f\). Note that by (5.2) it is sufficient to prove that \(u_n \rightharpoonup u\) in \(L^2((\eta ,T);H^{\alpha /2}(\mathbb {R}^d))\). Let us test the equation for \(u_n\) with \(\phi = u_n\), integrate over \((\eta ,T)\). Then:
By uniform boundedness of \(\Vert u_n \Vert _{L^2((\eta ,T)\times \mathbb {R}^d)}\), we conclude that also \(\Vert u_n \Vert _{L^{2}((\eta ,T);H^{\alpha /2}(\mathbb {R}^d))}\) is uniformly bounded, which implies \(u_n \rightharpoonup u\) in \(L^2((\eta ,T);H^{\alpha /2}(\mathbb {R}^d))\), as desired.
The desired result for \(P^{\Omega _n,\rho }_{\eta ,s}\) follows by the same argument, using (5.4) to conclude boundedness in \(L^2((\eta ,T);H^{\alpha /2}(\mathbb {R}^d))\). \(\square \)
Proof of Lemma 2.2
Let \(f \in L^2(\mathbb {R}^d)\) with \(f \ge 0\), and \(\Omega _n \subset \mathbb {R}^d\) as in Lemma 5.3. We proceed as in the proof of Proposition 4.4 in [30] to deduce for a.e. \(0 \le \eta< s < T\) and \(n \in \mathbb {N}\):
where \(K_{\rho }(y) = \sup _{z \in \mathbb {R}^d, t \in (\eta ,T)} k(t;z,y)\mathbbm {1}_{\{ \vert z-y \vert \ge \rho \}}(y)\). The proof of (5.5) directly follows by application of the parabolic maximum principle Lemma 5.2 for \(L_t\) to
where \(\phi _n \in C_c^{\infty }(\mathbb {R}^d)\) with \(0 \le \phi _n \le 1\) and \(\phi _n \equiv 1\) in \(\Omega _n\). Taking \(n \rightarrow \infty \) in (5.5) implies
where we used Lemma 5.3 and (2.6).
By (2.9) it is easy to deduce for a.e. \(0 \le \eta< s < T\) and a.e. \(x,y \in \mathbb {R}^d\):
(2.11) is now a direct consequence of (2.7) and \(k_{\le }\). The proof of (2.12) follows by carrying out the same arguments as in the proof of Proposition 4.4 in [30] with
where \(K_{\rho }'{:}{=} \sup _{(t,x) \in (0,T) \times \mathbb {R}^d} \int _{\mathbb {R}^d \setminus B_{\rho }(x)} k(t;x,y) \text {d}y\). This time, we apply the parabolic maximum principle Lemma 5.2 for \(L^{\rho }_t\) and obtain for a.e. \(0 \le \eta< s < T\) and a.e. \(x,y \in \mathbb {R}^d\):
Note that by Hölder regularity of weak solutions to (1.3), it is possible to obtain the desired results for every \(\eta ,s,x,y\). \(\square \)
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
Kassmann, M., Weidner, M. Upper heat kernel estimates for nonlocal operators via Aronson’s method. Calc. Var. 62, 68 (2023). https://doi.org/10.1007/s00526-022-02398-y
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s00526-022-02398-y