Abstract
Here, we consider periodic homogenization for time-fractional Hamilton–Jacobi equations. By using the perturbed test function method, we establish the convergence, and give estimates on a rate of convergence. A main difficulty is the incompatibility between the function used in the doubling variable method, and the non-locality of the Caputo derivative. Our approach is to provide a lemma to prove the rate of convergence without the doubling variable method with respect to the time variable, which is a key ingredient.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Let \(T > 0\) and \(\alpha \in (0, 1)\) be given constants and \(\varepsilon > 0\) be a parameter. We are concerned with time-fractional Hamilton–Jacobi equations:
Here, \(u^\varepsilon :{\mathbb {R}}^N \times [0,T] \rightarrow {\mathbb {R}}\) is an unknown function. The Hamiltonian \(H:{\mathbb {R}}^N \times {\mathbb {R}}^N \rightarrow {\mathbb {R}}\), the initial function \(u_0:{\mathbb {R}}^N \rightarrow {\mathbb {R}}\) are given continuous functions, which always satisfy
-
(A1)
The function H is uniformly coercive in the y-variable, i.e.,
$$\begin{aligned} \displaystyle \lim _{r \rightarrow \infty } \inf \{ H(y, p) \mid y \in {\mathbb {R}}^N, |p| \ge r \} = \infty . \end{aligned}$$ -
(A2)
The function \(y \mapsto H(y, p)\) is \({\mathbb {Z}}^N\)-periodic, i.e.,
$$\begin{aligned} H(y, p) = H(y+k, p) \quad \text{ for } \, \, y, p \in {\mathbb {R}}^N, k \in {\mathbb {Z}}^N. \end{aligned}$$ -
(A3)
\(u_0 \in \mathrm{Lip\,}({\mathbb {R}}^N) \cap \mathrm{BUC\,}({\mathbb {R}}^N)\), where we denote by \(\mathrm{Lip\,}({\mathbb {R}}^N)\) and \(\mathrm{BUC\,}({\mathbb {R}}^N)\), respectively, the sets of all Lipschitz continuous functions, and bounded, uniformly continuous functions on \({\mathbb {R}}^N\).
Moreover, we denote by \(\partial _{t}^{\alpha }u^\varepsilon \) the Caputo fractional derivative of \(u^\varepsilon \) with respect to t, that is,
for all \((x, t)\in {\mathbb {R}}^N \times (0,T)\), where \(\Gamma \) is the Gamma function. If we formally consider \(\alpha =1\), then the Caputo fractional derivative coincides with the normal derivative with respect to t, and thus (1.1) turns out to be the standard Hamilton–Jacobi equation.
Fractional derivatives attracted great interest from both mathematics and applications within the last few decades, and developed in wide fields (see [11, 14, 16] for instance). Studying differential equations with fractional derivatives is motivated by mathematical models that describe diffusion phenomena in complex media like fractals, which is sometimes called anomalous diffusion. It has inspired further research on numerous related topics.
The well-posedness of viscosity solutions to (1.1) was established by [9, 17, 19]. More precisely, a comparison principle, Perron’s method, and stability results have been established in [9, 17, 19]. We also mention that the equivalence of two weak solutions for linear uniformly parabolic equations with Caputo’s time-fractional derivatives was proved in [8] by using the resolvent type approximation introduced by [7]. In [12], the authors studied the regularity of viscosity solutions of (1.1), and got the large-time asymptotic result in some special settings.
In our paper, we are interested in the asymptotic behavior of \(u^\varepsilon \) as \(\varepsilon \rightarrow 0\). This singular limit problem is called “homogenization problem” with a background of the material science. Lions, Papanicolau and Varadhan in [13] were the first who started to study homogenization for Hamilton–Jacobi type equations, and after the perturbed test function method was introduced by Evans [5, 6], there has been much literature concerning on homogenization for nonlinear partial differential equations. It was proved in [13] that, when \(\alpha =1\), under assumptions (A1)–(A3), \(u^\varepsilon \) converges to \({\overline{u}}\) locally uniformly on \({\mathbb {R}}^N \times [0,T]\) as \(\varepsilon \rightarrow 0\), and \({\overline{u}}\) solves the effective equation (1.4) with \(\alpha =1\). The effective Hamiltonian \({\overline{H}} \in C({\mathbb {R}}^N)\) is determined in a nonlinear way by H as following. For each \(p \in {\mathbb {R}}^N\), it was shown in [13] that there exists a unique constant \({\overline{H}}(p)\in {\mathbb {R}}\), which is called the effective Hamiltonian, such that the stationary problem has a continuous viscosity solution to
If needed, we write \(v=v(y,p)\) to clearly demonstrate the nonlinear dependence of v on p. We call a problem to find a pair \((v(\cdot , p), {\overline{H}}(p)) \in \mathrm{Lip\,}({\mathbb {T}}^N) \times {\mathbb {R}}\) to satisfy (1.2) the cell problem. It is worth mentioning that in general v(y, p) is not unique even up to additive constants.
Heuristically, owing to the two-scale (i.e., x and \(\frac{x}{\varepsilon }\)) asymptotic expansion, of solutions \(u^\varepsilon \) to (1.1) of the form
by plugging this into (1.1), and using the coercivity of Hamiltonian, we can naturally expect that
Plugging this into (1.1) again, letting v be a solution of (1.2) with \(p= D_x{\overline{u}}(x, t)\), we can expect that \(u^\varepsilon \) converges to the limit function \({\overline{u}}\) as \(\varepsilon \rightarrow 0\) which is a solution to
Our main goal of this paper is to establish the convergence result of \(u^\varepsilon \) in \(C({\mathbb {R}}^N\times [0,T])\), and obtain a rate of convergence of \(u^\varepsilon \) to u, that is, an estimate for \(\Vert u^\varepsilon -u\Vert _{L^\infty ({\mathbb {R}}^N \times [0,T])}\) for any given \(T>0\) with resect to \(\varepsilon \). We give two main results.
Theorem 1.1
Let \(u^\varepsilon \) be the viscosity solution to (1.1). The function \(u^\varepsilon \) converges locally uniformly in \({\mathbb {R}}^N \times [0, T]\) to the function \({\overline{u}} \in C({\mathbb {R}}^N \times [0, T])\) as \(\varepsilon \rightarrow 0\), where \({\overline{u}}\) is the unique viscosity solution to (1.4).
Theorem 1.2
We additionally assume
-
(A4)
There exists \(C > 0\) such that
$$\begin{aligned} |H(x, p) - H(y, q)| \le C(|x - y| + |p - q|) \quad \text{ for } \, \, x, y, p, q \in {\mathbb {R}}^N. \end{aligned}$$ -
(A5)
There exists \(C > 0\) such that
$$\begin{aligned} H(x, p) \ge C^{-1}|p| - C \quad \text{ for } \, \, x, p \in {\mathbb {R}}^N. \end{aligned}$$
Let \(u^\varepsilon \) and \({\overline{u}}\) be the viscosity solutions of (1.1) and (1.4), respectively. For each \(\nu \in (0, 1)\), there exists a constant \(C > 0\) such that
By the standard perturbed test function method introduced by [5], it is not hard to obtain Theorem 1.1. To obtain Theorem 1.2, we use the method introduced by [3] which is a combination of the perturbed test function method and the discount approximation. A main difficulty we face here is the incompatibility between the doubling variable method which is often used in the theory of viscosity solutions and the non-locality of the Caputo derivative. More precisely, setting \(\varphi (t,s)=(t-s)^2\), usually we have \(\varphi _{t}(t,s)=-\varphi _{s}(t,s)\), which is an elementary, but important trick in the theory of viscosity solutions, however, we have
Our approach is to provide a lemma to prove Theorem 1.2 without using the doubling variable method with respect to time variable (see Lemma 5.2), which is our key ingredient of this paper. It is inspired by [9, Lemma 3.3].
The study of a rate of convergence in homogenization of Hamilton–Jacobi equations was started by Capuzzo-Dolcetta and Ishii [3], and in recent years, there has been much interest on the optimal convergence rate. Mitake et al. [15] was the first who established the optimal convergence rate for convex Hamilton–Jacobi equations with several conditional settings by using the representation formula for \(u^\varepsilon \) from optimal control theory, and weak KAM methods. Then, Cooperman [4] obtained a near optimal convergence rate \(|u^\varepsilon (x,t)-{\overline{u}}(x,t)|\le C{\varepsilon }\log (C+{\varepsilon ^{-1}}t)\) when \(n\ge 3\) for general convex settings by using a theorem of Alexander [1], originally proved in the context of first-passage percolation. Finally, Tran and Yu [21] has established the optimal rate \({\mathcal {O}}(\varepsilon )\) by using an argument of Burago [2], which concludes the study of this whole program in the convex setting. It is worth mentioning that the best known result on a rate of convergence on \(u^\varepsilon \) for the stationary Hamilton–Jacobi equation with the general nonconvex Hamiltonian is still \({\mathcal {O}}(\varepsilon ^\frac{1}{3})\) obtained by [3]. The argument in [3] can be easily adapted to (1.1) with \(\alpha =1\) (see [20, Theorem 4.37]). We also refer to [10, 18, 22] for other development on this subject. In all works [4, 10, 15, 18, 21, 22], the argument relies on the optimal control formula for \(u^\varepsilon \), and therefore it is seemingly rather challenging to obtain the optimal rate of convergence of (1.1) for \(0< \alpha < 1\), which remains completely open.
Organization of the paper. The paper is organized as follows. In Sect. 2, we recall the definition of viscosity solutions to (1.1), and give several basic results on time-fractional Hamilton–Jacobi equations in the theory of viscosity solutions. In Sect. 3, we give regularity results of viscosity solutions to (1.1), and (1.4). Sections 4 and 5 are devoted to prove the convergence of \(u^\varepsilon \), Theorem 1.1, and a convergence rate of \(u^\varepsilon \) to \({\overline{u}}\) in \(L^\infty \), Theorem 1.2.
2 Preliminaries
In this section, we recall the definition of viscosity solutions. First, we give several elementary facts of the Caputo derivative. These are well-known, but we give proofs to make the paper self-contained.
Proposition 2.1
Let \(f:[0, T] \mapsto {\mathbb {R}}\) be a function such that \(f \in C^1((0, T]) \cap C([0, T])\) and \({\partial _t f} \in L^1(0, T)\). Then,
for any \(t \in (0, T]\), where \({\tilde{f}}\) is defined by
Proof
For \(\varepsilon > 0\), using the integration by parts, we have
Note that
Thus,
Using the smoothness of f and taking the limit as \(\varepsilon \rightarrow 0\) above, we get (2.1).
On the other hand, using the change of variables, we obtain
which completes the proof. \(\square \)
From the observation in Proposition 2.1, we set
for \(t \in (0, T]\) and \(0 \le a < b \le t\).
Proposition 2.2
Let \(f \in C^1((0, T]) \cap C([0, T])\). For any \(t \in (0, T]\), the function \(K_{(0, t)}[f](t)\) exists and is continuous in (0, T].
Proof
For any \(s \in (0, T]\), we set
where \(\chi _I\) is the characteristic function, i.e., \(\chi _{I}(\tau ) = 1\) if \(\tau \in I\) and \(\chi _{I}(\tau ) = 0\) if \(\tau \notin I\) for an interval I. Fix \(t \in (0, T]\). It suffices to prove that \(g_s \rightarrow g_t\) in \(L^1(0, T)\) as \(s \rightarrow t\). Let \(\rho \in (0,\frac{t}{2})\), and for \(s \in (t - \rho , \min \{t + \rho , T \})\), we have
The right hand side is integrable on (0, T). Thus, using the dominated convergence theorem, we get the desired result. \(\square \)
Now, we define the definition of viscosity solutions to (1.1).
Definition 1
An upper semicontinuous function \(u^\varepsilon : {\mathbb {R}}^N \times [0, T] \mapsto {\mathbb {R}}\) is said to be a viscosity subsolution of (1.1) if for any \(\varphi \in C^1({\mathbb {R}}^N \times (0, T]) \cap C({\mathbb {R}}^N \times [0, T])\) one has
whenever \(u^\varepsilon - \varphi \) attains a maximum at \(({\hat{x}}, {\hat{t}}) \in {\mathbb {R}}^N \times (0, T]\) over \({\mathbb {R}}^N \times [0, T]\) and \(u^\varepsilon (\cdot , 0) \le u_0\) on \({\mathbb {R}}^N\).
Similarly, a lower semicontinuous function \(u^\varepsilon : {\mathbb {R}}^N \times [0, T] \mapsto {\mathbb {R}}\) is said to be a viscosity supersolution of (1.1) if for any \(\varphi \in C^1({\mathbb {R}}^N \times (0, T]) \cap C({\mathbb {R}}^N \times [0, T])\) one has
whenever \(u^\varepsilon - \varphi \) attains a minimum at \(({\hat{x}}, {\hat{t}}) \in {\mathbb {R}}^N \times (0, T]\) over \({\mathbb {R}}^N \times [0, T]\) and \(u^\varepsilon (\cdot , 0) \ge u_0\) on \({\mathbb {R}}^N\).
Finally, we call \(u^\varepsilon \in C({\mathbb {R}}^N \times [0, T])\) a viscosity solution of (1.1) if \(u^\varepsilon \) is both a viscosity subsolution and supersolution of (1.1).
We give several equivalent conditions for viscosity subsolutions/supersolutions to (1.1), which are often used in Sects. 3–5.
Proposition 2.3
For any \(\varepsilon > 0\), let \(u^\varepsilon : {\mathbb {R}}^N \times [0, T] \mapsto {\mathbb {R}}\) be an upper semicontinuous function. Then, the following statements are equivalent.
-
(a)
\(u^\varepsilon \) is a viscosity subsolution of (1.1).
-
(b)
For any \(\varphi \in C^1({\mathbb {R}}^N \times (0, T]) \cap C({\mathbb {R}}^N \times [0, T])\) one has
$$\begin{aligned} \displaystyle J[u^\varepsilon ]({\hat{x}}, {\hat{t}}) + K_{(0, \rho )}[\varphi ]({\hat{x}}, {\hat{t}}) + K_{(\rho , {\hat{t}})}[u^\varepsilon ]({\hat{x}}, {\hat{t}}) + H \left( \frac{{\hat{x}}}{\varepsilon }, D\varphi ({\hat{x}}, {\hat{t}}) \right) \le 0 \end{aligned}$$for \(0< \rho < {\hat{t}}\), whenever \(u^\varepsilon - \varphi \) attains a maximum at \(({\hat{x}}, {\hat{t}}) \in {\mathbb {R}}^N \times (0, T]\) over \({\mathbb {R}}^N \times [0, T]\) and \(u^\varepsilon (\cdot , 0) \le u_0\) on \({\mathbb {R}}^N\).
-
(c)
\(K_{(0, {\hat{t}})}[u^\varepsilon ]({\hat{x}}, {\hat{t}})\) exists and for any \(\varphi \in C^1({\mathbb {R}}^N \times (0, T]) \cap C({\mathbb {R}}^N \times [0, T])\) one has
$$\begin{aligned} \displaystyle J[u^\varepsilon ]({\hat{x}}, {\hat{t}}) + K_{(0, {\hat{t}})}[u^\varepsilon ]({\hat{x}}, {\hat{t}}) + H \left( \frac{{\hat{x}}}{\varepsilon }, D\varphi ({\hat{x}}, {\hat{t}}) \right) \le 0, \end{aligned}$$whenever \(u^\varepsilon - \varphi \) attains a local maximum at \(({\hat{x}}, {\hat{t}}) \in {\mathbb {R}}^N \times (0, T]\) over \({\mathbb {R}}^N \times [0, T]\) and \(u^\varepsilon (\cdot , 0) \le u_0\) on \({\mathbb {R}}^N\).
Similarly, we obtain equivalent conditions for viscosity supersolutions.
Proof
(a) \(\Rightarrow \) (b). Take \(\varphi \in C^1({\mathbb {R}}^N \times (0, T]) \cap C({\mathbb {R}}^N \times [0, T])\) such that \(u^\varepsilon - \varphi \) has a strict maximum at \(({\hat{x}}, {\hat{t}}) \in {\mathbb {R}}^N \times (0, T]\). Without loss of generality, we may assume that \((u^\varepsilon - \varphi )({\hat{x}}, {\hat{t}}) = 0\). Fix \(0< \rho < {\hat{t}}\). For any \(\sigma > 0\), take a function \(\varphi _\sigma \in C^1({\mathbb {R}}^N \times (0, T]) \cap C({\mathbb {R}}^N \times [0, T])\) satisfying
-
\(\varphi _\sigma = \varphi \) in \(B(({\hat{x}}, {\hat{t}}), \frac{\rho }{2})\),
-
\(u^\varepsilon \le \varphi _\sigma \le \varphi \) in \(B(({\hat{x}}, {\hat{t}}), \rho )\),
-
\(u^\varepsilon \le \varphi _\sigma \le u^\varepsilon + \sigma \) in \(\big ({\mathbb {R}}^N\times [0,T]\big )\setminus B(({\hat{x}}, {\hat{t}}), \rho )\).
Noting that \(\max (u^\varepsilon - \varphi _\sigma ) = (u^\varepsilon - \varphi _\sigma )({\hat{x}}, {\hat{t}})\), by (a), we have
It is clear that \(\lim _{\sigma \rightarrow 0} J[\varphi _\sigma ]({\hat{x}}, {\hat{t}}) = J[u^\varepsilon ]({\hat{x}}, {\hat{t}})\) and \(D\varphi _\sigma ({\hat{x}}, {\hat{t}}) = D\varphi ({\hat{x}}, {\hat{t}})\). Noting that \(\varphi ({\hat{x}}, {\hat{t}}) - \varphi _\sigma ({\hat{x}}, {\hat{t}}) \le \varphi ({\hat{x}}, {\hat{t}} - \tau ) - \varphi _\sigma ({\hat{x}}, {\hat{t}} - \tau )\) for \(0< \tau < \rho \), we have
By the dominated convergence theorem, we have
Thus, sending \(\sigma \rightarrow 0\) in (2.2), we obtain
(b) \(\Rightarrow \) (c). Assume that there exists \(\varphi \in C^1({\mathbb {R}}^N \times (0, T]) \cap C({\mathbb {R}}^N \times [0, T])\) such that \(u^\varepsilon - \varphi \) has a strict maximum at \(({\hat{x}}, {\hat{t}}) \in {\mathbb {R}}^N \times (0, T)\). By (b), we have
which implies that
for some \(C > 0\) which is independent of \(\rho > 0\). Noting that \(\varphi \in C^1({\mathbb {R}}^N \times (0, T])\), we have
For \(r \in {\mathbb {R}}\), we write \(r_+ \mathrel {\mathop :}=\max \{r, 0 \}\), \(r_- \mathrel {\mathop :}={-\min \{r, 0 \}}\). Then,
Noting that \(u^\varepsilon ({\hat{x}}, {\hat{t}}) - \varphi ({\hat{x}}, {\hat{t}}) \ge u^\varepsilon ({\hat{x}}, {\hat{t}} - \tau ) - \varphi ({\hat{x}}, {\hat{t}} - \tau )\), we obtain
for some \(C^\prime > 0\) which is independent of \(\rho > 0\). By the monotone convergence theorem, sending \(\rho \rightarrow 0\) in (2.3), we obtain
(c) \(\Rightarrow \) (a). Take \(\varphi \in C^1({\mathbb {R}}^N \times (0, T]) \cap C({\mathbb {R}}^N \times [0, T])\) such that \(u^\varepsilon - \varphi \) has a strict maximum at \(({\hat{x}}, {\hat{t}}) \in {\mathbb {R}}^N \times (0, T]\). Noting that \((u^\varepsilon -\varphi )({\hat{x}}, {\hat{t}}-\tau ) \le (u^\varepsilon -\varphi )({\hat{x}}, {\hat{t}})\) for all \(\tau \in [0, {\hat{t}}]\), we have
This inequality together with (c), we get (a). \(\square \)
We state basic results for time-fractional Hamilton–Jacobi equations, and we refer to [17] for the proofs.
Proposition 2.4
(Comparison principle, [9, Theorem 3.1]) Let \(u, v:{\mathbb {R}}^N \times [0, T] \rightarrow {\mathbb {R}}\) be a viscosity subsolution and a viscosity supersolution of (1.1), respectively. If \(u, v \in \mathrm{BUC\,}({\mathbb {R}}^N \times [0, T])\) and \(u(\cdot , 0) \le v(\cdot , 0)\) on \({\mathbb {R}}^N\), then \(u \le v\) on \({\mathbb {R}}^N \times [0, T]\).
Proposition 2.5
(Existence of a solution, [9, Theorem 4.2]) Let \(u_-, u_+:{\mathbb {R}}^N \times [0, T] \rightarrow {\mathbb {R}}\) be a viscosity subsolution and a viscosity supersolution of (1.1), respectively. Suppose that \(u_- \le u_+\) in \({\mathbb {R}}^N \times [0, T] \rightarrow {\mathbb {R}}\). There exists a viscosity solution u of (1.1) that satisfies \(u_- \le u \le u_+\) in \({\mathbb {R}}^N \times [0, T] \rightarrow {\mathbb {R}}\).
Proposition 2.5 follows from Lemmas 2.6 and 2.7. We denote \(S^-\) and \(S^+\) by the set of all subsolutions and supersolutions of (1.1), respectively.
Lemma 2.6
(Closedness under supremum/infimum operator, [9, Lemma 4.1]) Let X be a nonempty subset of \(S^-\) (resp., \(S^+\)). Set
If \(u < \infty \) (resp., \(u > -\infty \)) on \({\mathbb {R}}^N \times [0, T]\), then u is a viscosity subsolution (resp., supersolution) of (1.1).
Lemma 2.7
Let \(u_+:{\mathbb {R}}^N \times [0, T] \rightarrow {\mathbb {R}}\) be a supersolution of (1.1). Define
If \(u \in X\) is not a supersolution of (1.1), then there exists a function \(w \in X\) and a point \((y, s) \in {\mathbb {R}}^N \times (0, T]\) such that \(u(y, s) < w(y, s)\).
Proof
Since if u is not a supersolution of (1.1), there exists \((({\hat{x}}, {\hat{t}}), \varphi ) \in ({\mathbb {R}}^N \times (0, T]) \times (C^1({\mathbb {R}}^N \times (0, T]) \cap C({\mathbb {R}}^N \times [0, T]))\) such that
and
For \(r > 0\) small enough, we have
where we denote by \(B(({\hat{x}}, {\hat{t}}), 2r)\) a cylindrical neighborhood of \(({\hat{x}}, {\hat{t}})\).
It is clear to see that \(\varphi \le u \le u_+\) in \({\mathbb {R}}^N \times [0, T]\). If \((u_+ - \varphi )({\hat{x}}, {\hat{t}})=0\), then it is clear to see a contradiction since \(u_+\) is a supersolution. Thus, we only consider the case where \(\lambda \mathrel {\mathop :}=\frac{1}{2}(u_+ - \varphi )({\hat{x}}, {\hat{t}}) > 0\). Since \(u_+ - \varphi \) is lower semicontinuous, if \(r > 0\) is small enough, we have \(\varphi + \lambda \le u_+\) in \(B(({\hat{x}}, {\hat{t}}), 2r)\). Since \(u > \varphi \) in \({\mathbb {R}}^N \times [0, T] {\setminus } \{({\hat{x}}, {\hat{t}})\}\), there exists \(\lambda ^\prime \in (0, \lambda )\) such that \(\varphi + 2\lambda ^\prime \le u\) in \(\big (B({\hat{x}}, 2r) \times [0, T]\big ) {\setminus } B(({\hat{x}}, {\hat{t}}), r)\). Define the function \(w:{\mathbb {R}}^N \times [0, T] \mapsto {\mathbb {R}}\) by
We claim that \(w \in X\). It is easy to see that \(w \le u_+\) in \({\mathbb {R}}^N \times [0, T]\). Take \(\psi \in C^1({\mathbb {R}}^N \times (0, T]) \cap C({\mathbb {R}}^N \times [0, T])\) such that \(w - \psi \) has a strict maximum at \(({\hat{y}}, {\hat{s}}) \in {\mathbb {R}}^N \times (0, T]\) with \((w-\psi )({\hat{y}}, {\hat{s}})=0\). If \(w = u\) at \(({\hat{y}}, {\hat{s}})\), since u is a subsolution, we are done. We consider the case where \(w = \varphi + {\lambda '}\) at \(({\hat{y}}, {\hat{s}})\). By definition of w, we have \(({\hat{y}}, {\hat{s}}) \in B(({\hat{x}}, {\hat{t}}), r)\) and \(w({\hat{y}}, {\hat{s}} - \tau ) \ge \varphi ({\hat{y}}, {\hat{s}} - \tau ) + \lambda ^\prime \) for \(\tau \in [0, {\hat{s}}]\). Noting that \(0 = (w - \psi )({\hat{y}}, {\hat{s}}) \ge (w - \psi )({\hat{y}}, {\hat{s}} - \tau )\), we have
for \(\tau \in [0, {\hat{s}}]\) which yields \(J[\psi ]({\hat{y}}, {\hat{s}}) \le J[\varphi ]({\hat{y}}, {\hat{s}})\) and \(K_{(0, {\hat{s}})}[\psi ] \)\( ({\hat{y}}, {\hat{s}}) \le K_{(0, {\hat{s}})}[\varphi ]({\hat{y}}, {\hat{s}})\). Also, since \(\varphi {+\lambda '} = \psi \) at \(({\hat{y}}, {\hat{s}})\) and \(0 > w - \psi \ge \varphi + \lambda ^\prime - \psi \) in \(\big ({\mathbb {R}}^N \times [0, T]\big ) {\setminus } \{ ({\hat{y}}, {\hat{s}}) \}\), we have \(D\varphi = D\psi \) at \(({\hat{y}}, {\hat{s}})\). Combining this with (2.5), we obtain
which implies that \(w \in S^-\) and finishes the proof. \(\square \)
3 Regularity estimates
Proposition 3.1
Let \(u^\varepsilon \) be the viscosity solution of (1.1). There exists a constant \(M^\prime > 0\) depending only on \(u_0, H\) and T such that
Proof
We set \(u^+(x, t) \mathrel {\mathop :}=u_0(x) + M t^\alpha \) and \(u^-(x, t) \mathrel {\mathop :}=u_0(x) - M t^\alpha \) with
Then, \(u^+\) and \(u^-\) are a viscosity supersolution and a viscosity subsolution of (1.1), respectively. By Proposition 2.4, we obtain
on \({\mathbb {R}}^N \times [0, T]\). Thus, we choose \(M^{\prime } \mathrel {\mathop :}=\Vert u_0\Vert _{\infty } + M T^\alpha \). This completes the proof. \(\square \)
Proposition 3.2
Let \(u^\varepsilon \) be the viscosity solution of (1.1). There exists a constant \(M > 0\) depending only on \(u_0\) and H such that
for \((x, t, s) \in {\mathbb {R}}^N \times [0, T] \times [0, T]\).
Proof
Let \(u^+\) be the function defined in the proof of Proposition 3.1, and let
where we denote \(S^-\) by the set of all subsolutions of (1.1) and M is the constant defined by (3.2). Set \(u(x, t) \mathrel {\mathop :}=\sup _{v \in X} v(x, t)\) for \((x, t) \in {\mathbb {R}}^N \times [0, T]\). It is clear to see that u satisfies (3.3).
Next, let us prove that u is a viscosity solution of (1.1). Noting that \(u_0(x) - Mt^\alpha \in X\), by Lemma 2.6, we see that u is a viscosity subsolution of (1.1). Suppose that u is not a supersolution of (1.1). Then there exists \((({\hat{x}}, {\hat{t}}), \varphi ) \in ({\mathbb {R}}^N \times (0, T]) \times (C^1({\mathbb {R}}^N \times (0, T]) \cap C({\mathbb {R}}^N \times [0, T]))\) such that
and
By Lemma 2.7, there exists a viscosity subsolution \(w \in \mathrm{USC\,}({\mathbb {R}}^N \times [0, T])\) of (1.1) satisfying \(w \le u^+\) on \({\mathbb {R}}^N \times [0, T]\) and \(w({\hat{x}}, {\hat{t}}) > u({\hat{x}}, {\hat{t}})\). By construction of w in the proof of Lemma 2.7, it is clear to see that w satisfies (3.3), and also \(w \in X\). This immediately gives a contradiction. By the uniqueness of viscosity solutions of (1.1), we obtain \(u = u^\varepsilon \) in \({\mathbb {R}}^N \times [0, T]\), which completes the proof. \(\square \)
Proposition 3.3
Let \(u^\varepsilon \) be the viscosity solution of (1.1). There exists a modulus of continuity \(\omega \in C([0, \infty ))\) independent of \(\varepsilon \) such that
Furthermore, assume (A5). There exists a constant \(C > 0\) independent of \(\varepsilon \) such that
for \((x, y, t) \in {\mathbb {R}}^N \times {\mathbb {R}}^N \times [0, T]\). In particular, for each \(\nu \in (0, 1)\), there exists a constant \(C > 0\) independent of \(\varepsilon \) such that
We do not know whether the regularity 3.4 is optimal or not. It comes from the nonlocal operator \(K_{(0,t)}[u]\) as seen in the proof below. We follow the proof of [12, Theorem 2.2] with a slight modification.
Definition 2
Let \(f: [0, T] \mapsto {\mathbb {R}}\) be a bounded function. For each \(\delta > 0\), we define \(f^\delta \) and \(f_\delta \) by
for \(t \in [0, T]\). We call \(f^\delta \) and \(f_\delta \) the sup-convolution and the inf-convolution, respectively.
Lemma 3.4
Let \(f: [0, T] \mapsto {\mathbb {R}}\) be a bounded function. There exists \(C > 0\), for each \(\delta > 0\),
-
(a)
\(f(t) \le f^\delta (t) \le \Vert f \Vert _{\infty }\) for \(t \in [0, T]\),
-
(b)
\(\displaystyle |f^\delta (t) - f^\delta (s)| \le \frac{C}{\delta } |t-s|\) for \(t, s \in [0, T]\).
Furthermore, assume that \(f \in C^{0, \alpha }([0, T])\) for \(\alpha \in (0, 1]\), then
-
(c)
\(|t - \xi _\delta |^{2-\alpha } \le 2 M \delta \) for \(\xi _\delta \in \mathop {\mathrm {arg\,max}}\limits \{ f(\xi ) - \frac{|t - \xi |^2}{2\delta } \mid \xi \in [0, T] \}\),
-
(d)
\(\Vert f^\delta - f \Vert _{\infty } \le (2\,M)^{\frac{2}{2 - \alpha }} \delta ^{\frac{\alpha }{2-\alpha }}\),
-
(e)
\(f^\delta \in C^{0, \alpha }([0, T])\). In particular, there exists \(C>0\) which only depends on M (is independent of \(\delta >0\)) such that \(\Vert f^\delta \Vert _{C^{0, \alpha }([0, T])}\le C\).
where M is the Hölder constant of f.
Proof
We omit the proofs of (a) and (b), as they are standard. We only prove (c), (d) and (e). Take \(\xi _\delta \in [0, T]\) so that \(f^\delta (t) = f(\xi _\delta ) - \frac{|t - \xi _\delta |^2}{2\delta }\). Note that
Thus, \(|t - \xi _\delta |^{2-\alpha } \le 2 M \delta \). By (3.5), we have
for \(t \in [0, T]\), which proves (d). Finally, let \(t, s \in [0, T]\). We consider the case \(|t - s|^{2 - \alpha } \ge \delta \). By applying (d), we have
We consider the case \(|t - s|^{2 - \alpha } \le \delta \). We have
By (c), we obtain
which completes the proof. \(\square \)
Lemma 3.5
Let u be a bounded viscosity subsolution of (1.1), and for \(\delta \in (0, 1)\), we write \(u^\delta \) and \(u_\delta \) for the sup-convolution and inf-convolution of u, respectively. Then \(u^\delta \) and \(u_\delta \) are, respectively, a viscosity subsolution and a viscosity supersolution of
where M is a constant such that \(M^2 \ge 2\Vert u \Vert _\infty \) and \(\eta _\delta \) is a constant such that \(\eta _\delta \rightarrow 0\) as \(\delta \rightarrow 0\).
We refer to [19, Lemma 4.2] and [17, Proposition 4.2] for similar results.
Proof
We only prove that \(u^\delta \) is a viscosity subsolution of (3.6). Take \((({\hat{x}}, {\hat{t}}), \varphi ) \in ({\mathbb {R}}^N \times (M\delta ^{\frac{1}{4}}, T]) \times (C^1({\mathbb {R}}^N \times (0, T]) \cap C({\mathbb {R}}^N \times [0, T]))\) so that
Let \({\hat{t}}_\delta \in [0, T]\) be such that
Then, we see that
Thus, we have \({\hat{t}}_\delta \ge {\hat{t}} - (2 \Vert u \Vert _\infty \delta )^\frac{1}{2}> M\delta ^\frac{1}{4} - (2 \Vert u \Vert _\infty \delta )^\frac{1}{2} \ge M\delta ^\frac{1}{4}(1 - \delta ^\frac{1}{4})> 0\). Noting that \((u^\delta - \varphi )(x, t) \le (u^\delta - \varphi )({\hat{x}}, {\hat{t}})\), we see that
for (x, t) in a neighborhood of \(({\hat{x}}, {\hat{t}})\). Thus, the function \((x, t) \mapsto u(x, t) - \varphi (x, t - {\hat{t}}_\delta + {\hat{t}})\) attains a local maximum at \(({\hat{x}}, {\hat{t}}_\delta )\). Therefore, there exists \({\tilde{\varphi }} \in C^1({\mathbb {R}}^N \times (0, T]) \cap C({\mathbb {R}}^N \times [0, T])\) such that \({\tilde{\varphi }}(x, t) = \varphi (x, t - {\hat{t}}_\delta + {\hat{t}})\) for (x, t) in a neighborhood of \(({\hat{x}}, {\hat{t}}_\delta )\) and
Since u is a viscosity subsolution to (1.1), by Proposition 2.3, we have
Define
Then, we see that
We set \(\Psi (\xi ) \mathrel {\mathop :}=u^\delta ({\hat{x}}, {\hat{t}}) + \frac{|{\hat{t}}_\delta - {\hat{t}}|^2}{\delta } - {\tilde{u}}({\hat{x}}, \xi + {\hat{t}}_\delta - {\hat{t}})\) for \(\xi \in (-\infty , {\hat{t}}]\). We divide into three cases: (1) \(\xi \le -M\delta ^\frac{1}{2}\), (2) \(-M\delta ^\frac{1}{2} < \xi \le M\delta ^\frac{1}{2}\), and (3) \(M\delta ^\frac{1}{2} < \xi \le {\hat{t}}\). First, we consider case (1). Since \(\xi + {\hat{t}}_\delta - {\hat{t}} \le -M\delta ^\frac{1}{2} + M\delta ^\frac{1}{2} = 0\), we have \({\tilde{u}}({\hat{x}}, \xi + {\hat{t}}_\delta - {\hat{t}}) = u({\hat{x}}, 0) \le u^\delta ({\hat{x}}, 0) = {\tilde{u}}^\delta ({\hat{x}}, \xi )\), where \({\tilde{u}}^\delta \) be the sup-convolution of \({\tilde{u}}\). Note that by Lemma 3.4 (c), in case (1), letting \(\tau _\delta \in \arg \max \left\{ {\tilde{u}}({\hat{x}},\tau )-\frac{(\xi -\tau )^2}{\delta }\right\} \), we have
if \(\delta >0\) is small enough. Thus, we have \({\tilde{u}}^\delta ({\hat{x}}, \xi )=u^\delta ({\hat{x}}, 0)\). We obtain
Next, in case (2), we have
Finally, we consider case (3). Since \(\xi + {\hat{t}}_\delta - {\hat{t}} > M\delta ^\frac{1}{2} - M\delta ^\frac{1}{2} = 0\), we have \({\tilde{u}}({\hat{x}}, \xi + {\hat{t}}_\delta - {\hat{t}}) = u({\hat{x}}, \xi + {\hat{t}}_\delta - {\hat{t}})\). Therefore, we obtain
Hence, we obtain
Noting that \({\hat{t}} > M\delta ^\frac{1}{4}\) and \(\frac{1}{2} \delta ^{\frac{1}{4}} > \delta ^{\frac{1}{2}}\) for small \(0< \delta < 1\), we have
which implies
where
This completes the proof. \(\square \)
Proof of Proposition 3.3
For \(\delta > 0\), we denote by \(u^{\varepsilon , \delta }\) the sup-convolution of \(u^\varepsilon \). For fixed \(y \in {\mathbb {R}}^N, t_0 > 0,\) and \(\beta , L > 0\), we set
where L will be fixed later. Since \(\Phi (x, t) \rightarrow -\infty \) as \(|x| \rightarrow \infty \) for any \(t \in [0, T]\) and \(\Phi \) is bounded from above, \(\Phi \) attains a maximum at a point \(({\hat{x}}, {\hat{t}}) \in {\mathbb {R}}^N \times [0, T]\). We claim that if L is sufficiently large, then \({\hat{x}} = y\). Suppose that \({\hat{x}} \ne y\). Since \(u^{\varepsilon , \delta }\) is the viscosity subsolution of (1.1), by Proposition 2.3, for every \(0< \rho < {\hat{t}}\), we have
By Proposition 3.2 and Lemma 3.4 (e), we have
and
Next, by Lemma 3.4 (b), we have
Combining these with (3.7), we obtain
By setting \(\rho = {\hat{t}}\delta ^{\frac{1}{1-\alpha }}\), we obtain
If we take \(L = L(\delta )\) large, then, by the coercivity of H, we get a contradiction.
Therefore, we have \(\Phi (x, {\hat{t}}) \le \Phi ({\hat{x}}, {\hat{t}}) = \Phi (y, {\hat{t}})\), i.e.,
Taking \(\beta \rightarrow 0\), we have
Therefore, by Lemma 3.4 (d), we obtain
for some \(C > 0\).
Finally, define
We claim that \(\omega (r) \rightarrow 0\) as \(r \rightarrow 0\). Letting \(\delta _r > 0\) so that \(L(\delta _r) = r^{-\frac{1}{2}}\), we have
Since \(\delta _r \rightarrow 0\) as \(r \rightarrow 0\), taking \(r \rightarrow 0\) as above, this completes the proof of modulus of continuity.
In addition, we assume (A5), i.e., there exists a constant \(C > 0\) such that
By (3.8), we have
By similar argument to the above, we obtain
for \(x, y \in {\mathbb {R}}^N\) and \(t \in (0, T]\). Thus, taking the infimum with respect to \(\delta \), we have
which completes the proof. \(\square \)
Remark 1
In [9, Lemma 6.1], they show that Lipschitz continuous with respect to the space variable. However, applying this result directly to our problem, we obtain
Therefore, this does not give us a uniform Lipschitz estimate. On the other hand, by [9, Lemma 6.1] we see that the limit problem (1.4) has the unique Lipschitz continuous viscosity solution. Therefore, we can expect a \(\varepsilon \)-uniform Lipschitz estimate, but this still remains open.
4 Perturbed test function method
Lemma 4.1
For each \(\varepsilon > 0\), let \(u^\varepsilon \) be the viscosity solution of (1.1). Set
Then, \(u^*\) and \(u_*\) are a viscosity subsolution and a viscosity supersolution of (1.4), respectively.
Proof
We only prove that \(u^*\) is a viscosity subsolution of (1.4), we can similarly prove that \(u_*\) is a viscosity supersolution of (1.4).
Let \(\varphi \in C^1({\mathbb {R}}^N \times (0, T]) \cap C({\mathbb {R}}^N \times [0, T])\) such that \(u^* - \varphi \) has a strict maximum at \(({\hat{x}}, {\hat{t}}) \in {\mathbb {R}}^N \times (0, T]\). Let \(P = D\varphi ({\hat{x}}, {\hat{t}}) \in {\mathbb {R}}^N\), and let \(v \in \) Lip \(({\mathbb {T}}^N)\) be a viscosity solution of (1.2). Choose a sequence \(\varepsilon _m \rightarrow 0\) \((m \in {\mathbb {N}})\) such that
We set
For every \(m \in {\mathbb {N}}\) and \(a > 0\), there exists \((x_{m, a}, y_{m, a}, t_{m, a}) \in {\mathbb {R}}^N \times {\mathbb {R}}^N \times (0, T)\) such that
and up to passing some subsequences
Notice that the function \((x, t) \mapsto \Phi ^{m, a}(x, y_{m, a}, t)\) has a maximum at \((x_{m, a}, t_{m, a})\). Since \(u^{\varepsilon _m}\) is a viscosity subsolution of (1.1), we have
On the other hand, notice that the function \(y \mapsto -\frac{1}{\varepsilon _m} \Phi ^{m, a}(x_{m, a}, \varepsilon _m y, t_{m, a})\) has a minimum at \(\frac{y_{m, a}}{\varepsilon _m}\). Since v is a viscosity supersolution of (1.2), we have
Since \(\Phi ^{m, a}(x_{m, a}, y_{m, a}, t_{m, a}) \ge \Phi ^{m, a}(x_{m, a}, x_{m, a}, t_{m, a})\), we have
where L is a Lipschitz constant of v. Thus, we have \( \frac{|x_{m, a} - y_{m, a}|}{a} \le 2\,L \). If necessary, taking a subsequence, we can assume that
By Proposition 2.2 and taking the limit \(a \rightarrow 0\) in (4.1) and (4.2) to get
Combining the above two inequality, and noting that \(|Q_m|\) is bounded, we obtain
for a modulus of continuity of H. Taking the limit \(m \rightarrow \infty \) to conclude that
\(\square \)
Lemma 4.2
Let \(u^*\) and \(u_*\) be the functions defined by Lemma 4.1. Then, \(u^* = u_* = {\overline{u}}\) in \({\mathbb {R}}^N \times [0, T]\), where \({\overline{u}}\) is the unique viscosity solution to (1.4).
Proof
By definition of half relaxed limit, we have \(u_* \le u^*\) in \({\mathbb {R}}^N \times [0, T]\). On the other hand, by the comparison principle for (1.4), we have \(u^* \le u_*\) in \({\mathbb {R}}^N \times [0, T]\), which implies the conclusion. \(\square \)
Theorem 1.1 is a straightforward result of Lemmas 4.1 and 4.2
5 Rate of convergence
Let us consider the discounted approximation for cell problem (1.2):
We recall some results of \(v^\lambda \), and refer to [3, Lemma 2.3] and [20, Lemma 4.38] for proofs.
Lemma 5.1
-
(a)
There exists a constant \(C > 0\) independent of \(\lambda > 0\) such that
$$\begin{aligned} \lambda |v^\lambda (y, p) - v^\lambda (y, q)| \le C|p - q| \quad \text{ for } \, \, y, p, q \in {\mathbb {R}}^N. \end{aligned}$$In particular, \(|{\overline{H}}(p) - {\overline{H}}(q)| \le C|p - q|\) for \(p, q \in {\mathbb {R}}^N\).
-
(b)
For any \(p \in {\mathbb {R}}^N\), there exists a constant \(C > 0\) independent of \(\lambda > 0\) and p such that
$$\begin{aligned} |\lambda v^\lambda (y, p) + {\overline{H}}(p)| \le C (1 + |p|) \lambda \quad \text{ for } \, \, y \in {\mathbb {R}}^N. \end{aligned}$$
The next lemma is a key ingredient to prove Theorem 1.2.
Lemma 5.2
For \(\varepsilon , \lambda > 0\), let \(u^\varepsilon \), \({\overline{u}}\) and \(v^\lambda \) be the viscosity solutions of (1.1), (1.4) and (5.1), respectively. Let \(\Phi : {\mathbb {R}}^{2N}\times [0, T] \mapsto {\mathbb {R}}\) be the function defined by
where \(\lambda = \varepsilon ^\theta \) and \(\theta , \beta , \delta \in (0, 1),\) and \(R > 0\). Assume that \(\Phi (x, y, t)\) attains a maximum at \(({\hat{x}}, {\hat{y}}, {\hat{t}}) \in {\mathbb {R}}^{2N} \times (0, T]\) over \({\mathbb {R}}^{2N} \times [0, T] \) and \(\delta ^{\frac{1}{2}} < \lambda \). There exists \(\varepsilon _0 \in (0, 1)\) such that for each \(\nu \in (0, 1)\), there exists \(C > 0\) which is independent of \(\varepsilon \) such that
for all \(\varepsilon \in (0, \varepsilon _0)\).
Proof
STEP 1. We claim that if \(1 - \theta - \beta > 0\), then there exists \(C > 0\) such that
Additionally, if \(\delta < \frac{1}{2}\), then
Noting that \(\Phi (0, 0, {\hat{t}}) \le \Phi ({\hat{x}}, {\hat{y}}, {\hat{t}})\), i.e.,
we have
Noting that, by Lemma 5.1,
we obtain
With \(\lambda = \varepsilon ^\theta \), by using the Young inequality, we obtain
which implies (5.2).
Next, by \(\Phi ({\hat{x}}, {\hat{x}}, {\hat{t}}) \le \Phi ({\hat{x}}, {\hat{y}}, {\hat{t}})\), i.e.,
and Proposition 3.3, Lemma 5.1, for \(\nu \in (0, 1)\), we obtain
We notice that \(\nu \) can be an arbitrary fixed number in (0, 1) in light of Proposition 3.3. It is sufficient to consider case \(|{\hat{x}} - {\hat{y}}| < 1\) if \(\varepsilon \) is small enough. Combining (5.4) with (5.2), we have
Since \(\delta < \frac{1}{2}\) and \(0 < 1 - \theta - \beta \), we get (5.3).
STEP 2. For \(\gamma > 0\), we consider the auxiliary function \(\Psi :{\mathbb {R}}^{4N}\times [0, T]^2 \mapsto {\mathbb {R}}\) defined by
Since \(\Psi (x, y, z, \xi , t, s) \rightarrow -\infty \) as \(|x|, |y|, |z|, |\xi | \rightarrow \infty \) for any \((t, s) \in [0, T]^2\) and \(\Psi \) is bounded from above, \(\Psi \) attains a maximum at a point \((x_\gamma , y_\gamma , z_\gamma , \xi _\gamma , t_\gamma , s_\gamma ) \in {\mathbb {R}}^{4N}\times [0, T]^2\). By the standard argument of the theory of viscosity solutions, we obtain
If necessary, taking \(\gamma \) sufficiently small, we can assume that \(t_\gamma , s_\gamma > 0\) because we assume \({\hat{t}}>0\). Our goal in STEP 2 is to obtain
where \(E_\gamma \) is a constant which satisfies \(E_\gamma \rightarrow 0\) as \(\gamma \rightarrow 0\), and will be defined later.
STEP 2-1. We claim that there exists \(C > 0\) such that
for some \(E_{1, \gamma } > 0\) satisfying \(E_{1, \gamma } \rightarrow 0\) as \(\gamma \rightarrow 0\). Notice that the function \((x, t) \mapsto \Psi (x, y_\gamma , z_\gamma , \xi _\gamma , t, s_\gamma )\) has a maximum at \((x_\gamma , t_\gamma )\). Since \(u^\varepsilon \) is a viscosity subsolution of (1.1), by Proposition 2.3, we have
On the other hand, notice that the function \(\xi \mapsto - \frac{1}{\varepsilon } \Psi (x_\gamma , y_\gamma , z_\gamma , \xi , t_\gamma , s_\gamma )\) has a minimum at \(\xi _\gamma \). Since \(v^\lambda \) is a viscosity supersolution of (5.1), we obtain
By \(\Psi (x_\gamma , y_\gamma , x_\gamma - y_\gamma , \xi _\gamma , t_\gamma , s_\gamma ) \le \Psi (x_\gamma , y_\gamma , z_\gamma , \xi _\gamma , t_\gamma , s_\gamma )\) and Lemma 5.1, we see that
which implies
Combining (5.7) with (5.8) and (5.9), by Lemma 5.1, we obtain
STEP 2-2. Next, we claim that there exists \(C > 0\) such that
for some \(E_{2, \gamma } > 0\) satisfying \(E_{2, \gamma } \rightarrow 0\) as \(\gamma \rightarrow 0\). Notice that \((y, s) \mapsto - \Psi (x_\gamma , y, z_\gamma , \xi _\gamma , t_\gamma , s)\) has a minimum at \((y_\gamma , s_\gamma )\). By Proposition 2.3, we have
Thus, by (5.2), (5.9) and Lemma 5.1, we obtain
Take \(0< \delta < \frac{1}{2}\) satisfying \(\delta ^{\frac{1}{2}} < \lambda = \varepsilon ^\theta \) to get (5.10). Combining (5.6) with (5.10), we get
We set \(E_\gamma \mathrel {\mathop :}=E_{1, \gamma } + E_{2, \gamma }\). This completes STEP 2.
STEP 3. Finally, we claim that taking the limit infimum \(\gamma \rightarrow 0\) in (5.5) yields that
We set
for \(0< r < \min \{t_\gamma , s_\gamma \}\). Notice that \(I_1 + \frac{\alpha }{\Gamma (1 - \alpha )}(I_2 + I_3) = J[u^\varepsilon ](x_\gamma , t_\gamma ) - J[{\overline{u}}](y_\gamma , s_\gamma ) + K_{(0, t_\gamma )}[u^\varepsilon ] (x_\gamma , t_\gamma ) - K_{(0, s_\gamma )}[{\overline{u}}](y_\gamma , s_\gamma )\).
First, by the continuity of \(u^\varepsilon \) and \({\overline{u}}\), we have
Next, by \(\Psi (x_\gamma , y_\gamma , z_\gamma , \xi _\gamma , t_\gamma , s_\gamma ) \ge \Psi (x_\gamma , y_\gamma , z_\gamma , \xi _\gamma , t_\gamma - \tau , s_\gamma - \tau )\) for all \(\tau \in [0, r]\), i.e.,
we have
Finally, we see that
where \(\chi _I\) is the characteristic function. Notice that by Proposition 3.2
Since the right-hand side is integrable on [0, T], by Fatou’s lemma, we obtain
Combining (5.11), (5.12) with (5.13), and sending \(\gamma \rightarrow 0\) in (5.5), by (5.3) we obtain
Note that as in the proof of (2.4) in Proposition 2.3, we can prove
for some \(C>0\) independent of r. Since \(C_r \rightarrow 0\) as \(r \rightarrow 0\) by (5.12), we obtain the desired result. \(\square \)
Proof of Theorem 1.2.
Let \(v^\lambda \) be the viscosity solution of (5.1). We consider the auxiliary function \(\Phi : {\mathbb {R}}^{2N}\times [0, T] \mapsto {\mathbb {R}}\) defined in Lemma 5.2, i.e.,
where \(\lambda = \varepsilon ^\theta , \theta , \beta \in (0, 1), 0< \delta < \lambda ^2\) and \(R > 0\). Noting that \(\Phi \) is a bounded from above and \(\Phi (x, y, t) \rightarrow -\infty \) as \(|x|, |y| \rightarrow \infty \) for any \(t \in [0, T]\), we see that \(\Phi \) attains a maximum at \(({\hat{x}}, {\hat{y}}, {\hat{t}}) \in {\mathbb {R}}^{2N}\times [0, T]\).
First, we set \(R:= R^\prime (\varepsilon ^{\theta - \frac{\beta (1-\nu )}{2-\nu }}+ \varepsilon ^{1-\theta -\beta })\). We claim that if \(R^\prime \) is sufficiently large, then \({\hat{t}} = 0\). Suppose that \({\hat{t}} > 0\). By Lemma 5.2, we have
By \(\Phi ({\hat{x}}, {\hat{y}}, 0) \le \Phi ({\hat{x}}, {\hat{y}}, {\hat{t}})\) and \(\Phi ({\hat{x}}, {\hat{y}}, {\hat{t}} - \tau ) \le \Phi ({\hat{x}}, {\hat{y}}, {\hat{t}})\) for all \(\tau \in [0, {\hat{t}}]\), we obtain
Combining two inequalities above with (5.14), we have
Thus, we set \(R^\prime > C \Gamma (1 - \alpha ) \Gamma (1 + \alpha )\), which implies a contradiction.
By a similar argument to the proof of Lemma 5.2, we have \(|{\hat{x}} - {\hat{y}}| \le C \varepsilon ^\frac{\beta }{2-\nu }\) if \(\delta < \frac{1}{2}\) and \(0< \theta < 1 - \beta \).
Next, by \(\Phi (x, x, t) \le \Phi ({\hat{x}}, {\hat{y}}, {\hat{t}}) = \Phi ({\hat{x}}, {\hat{y}}, 0)\) for all \((x, t) \in {\mathbb {R}}^N \times (0, T)\), we obtain, by Lemma 5.1,
for all \((x, t) \in {\mathbb {R}}^N \times (0, T)\). Therefore, by the optimal choice of parameters \(\theta = \beta = \frac{1}{3}\), and sending \(\delta \rightarrow 0\), we obtain
By a symmetric argument, we get the desired result. \(\square \)
Data availability
Data sharing is not applicable to this article as no new data were created in this study.
References
Alexander, K.. S.: Approximation of subadditive functions and convergence rates in limiting-shape results. Ann. Probab. 25(1), 30–55 (1997)
Burago, D.: Periodic metrics. Adv. Sov. Math. 9, 205–210 (1992)
Capuzzo-Dolcetta, I., Ishii, H.: On the rate of convergence in homogenization of Hamilton–Jacobi equations. Indiana Univ. Math. J. 50(3), 1113–1129 (2001)
Cooperman, W.: A near-optimal rate of periodic homogenization for convex Hamilton–Jacobi equations. Arch. Ration. Mech. Anal. 245(2), 809–817 (2022)
Evans, L.C.: The perturbed test function method for viscosity solutions of nonlinear PDE. Proc. R. Soc. Edinb. Sect. A 111(3–4), 359–375 (1989)
Evans, L.C.: Periodic homogenisation of certain fully nonlinear partial differential equations. Proc. R. Soc. Edinb. Sect. A 120(3–4), 245–265 (1992)
Giga, Y., Liu, Q., Mitake, H.: Discrete schemes for time-fractional fully nonlinear evolution equations and their convergence. Asymptot. Anal. 120(1–2), 151–162 (2020)
Giga, Y., Mitake, H., Sato, S.: On the equivalence of viscosity solutions and distributional solutions for the time-fractional diffusion equation. J. Differ. Equ. 316, 364–386 (2022)
Giga, Y., Namba, T.: Well-posedness of Hamilton–Jacobi equations with Caputo’s time fractional derivative. Commun. Partial Differ. Equ. 42(7), 1088–1120 (2017)
Han, Y. Jang, J.: Rate of Convergence in Periodic Homogenization for Convex Hamilton–Jacobi Equations with Multiscales. In preparation. arXiv:2303.05738
Kubica, A., Ryszewska, K., Yamamoto, M.: Time-Fractional Differential Equations—A Theoretical Introduction. SpringerBriefs in Mathematics, Springer, Singapore (2020)
Ley, O., Topp, E., Yangari, M.: Some results for the large time behavior of Hamilton–Jacobi Equations with Caputo time derivative. Discrete Contin. Dyn. Syst. Ser. A 41(8), 3555–3577 (2021)
Lions, P.-L., Papanicolaou, G., Varadhan, S.R.S.: Homogenization of Hamilton–Jacobi Equations, unpublished work (1987)
Metzler, R., Klafter, J.: The random walk’s guide to anomalous diffusion: a fractional dynamics approach. Phys. Rep. 339, 1–77 (2000)
Mitake, H., Tran, H.V., Yu, Y.: Rate of convergence in periodic homogenization of Hamilton–Jacobi equations: the convex setting. Arch. Ration. Mech. Anal. 233(2), 901–934 (2019)
Nakagawa, J., Sakamoto, K., Yamamoto, M.: Overview to mathematical analysis for fractional diffusion equations-new mathematical aspects motivated by industrial collaboration. J. Math. Ind. 2, 99–108 (2010)
Namba, T.: On existence and uniqueness of viscosity solutions for second order fully nonlinear PDEs with Caputo time fractional derivatives. NoDEA Nonlinear Differ. Equ. Appl. 25(3), 23 (2018)
Nguyen-Tien, H.: Optimal Convergence Rate for Homogenization of Convex Hamilton–Jacobi Equations in the Periodic Spatial-Temporal Environment. arXiv:2212.14782
Topp, E., Yangari, M.: Existence and uniqueness for parabolic problems with Caputo time derivative. J. Differ. Equ. 262(12), 6018–6046 (2017)
Tran, H.V.: Hamilton–Jacobi Equations: Theory and Applications. Graduate Studies in Mathematics, vol. 213. American Mathematical Society, Providence (2021)
Tran, H.V., Yu, Y.: Optimal Convergence Rate for Periodic Homogenization of Convex Hamilton–Jacobi Equations. arXiv:2112.06896
Tu, S.N.T.: Rate of convergence for periodic homogenization of convex Hamilton–Jacobi equations in one dimension. Asymptot. Anal. 121(2), 171–194 (2021)
Acknowledgements
The authors would like to thank Olivier Ley for helpful comments and suggestions. The authors would like to thank anonymous referees for carefully reading the first manuscript and giving them valuable comments.
Funding
Open access funding provided by The University of Tokyo. HM was partially supported by the JSPS grants: KAKENHI #22K03382, #21H04431, #20H01816, #19K03580, #19H00639. SS was supported by the Leading Graduate Course for Frontiers of Mathematical Sciences and Physics, The University of Tokyo, MEXT.
Author information
Authors and Affiliations
Contributions
This is a truly joint work.
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no conflict of interest.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
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
Mitake, H., Sato, S. On the rate of convergence in homogenization of time-fractional Hamilton–Jacobi equations. Nonlinear Differ. Equ. Appl. 30, 68 (2023). https://doi.org/10.1007/s00030-023-00880-w
Received:
Revised:
Accepted:
Published:
DOI: https://doi.org/10.1007/s00030-023-00880-w