Linear integro-differential operator and Open problems: Difference between pages

From nonlocal pde
(Difference between pages)
Jump to navigation Jump to search
imported>Xavi
 
imported>Luis
No edit summary
 
Line 1: Line 1:
The linear integro-differential operators that we consider ''in this wiki'' are the generators of [[Levy processes]]. According to the Levy-Kintchine formula, they have the general form
== Well posedness of the supercritical [[surface quasi-geostrophic equation]] ==
Let $\theta_0 : \R^2 \to \R$ be a smooth function either with compact support or periodic. Let $s \in (0,1/2)$. Is there a global classical solution $\theta :\R^2 \to \R$ for the SQG equation?
\begin{align*}
\theta(x,0) &= \theta_0(x) \\
\theta_t + u \cdot \nabla \theta &= 0 \qquad \text{in } \R^2 \times (0,+\infty)
\end{align*}
where $u = R^\perp \theta$ and $R$ stands for the Riesz transform.


\[ Lu(x) = \mathrm{tr} \, A(x) \cdot D^2 u + b(x) \cdot \nabla u + c(x) u + d(x) + \int_{\R^n} (u(x+y) - u(x) - y \cdot \nabla u(x) \chi_{B_1}(y)) \, \mathrm{d} \mu_x(y) \]
This is a very difficult open problem. It is believed that a solution would be a major step towards the understanding of Navier-Stokes equation. In the supercritical regime $s\in (0,1/2)$, the effect if the drift term is larger than the diffusion in small scales. Therefore, it seems unlikely that a proof of well posedness could be achieved with the methods currently known and listed in this wiki.
where $A(x)$ is a nonnegative matrix for all $x$, and $\mu_x$ is a nonnegative measure for all $x$ satisfying
\[ \int_{\R^n} \min(y^2 , 1) \mathrm{d} \mu_x(y) < +\infty. \]


The above definition is very general. Many theorems, and in particular regularity theorems, require extra assumptions in the kernels $K$. These assumptions restrict the study to certain sub-classes of linear operators. The simplest of all is the [[fractional Laplacian]]. We list below several extra assumptions that are usually made.
Note that if the relation between $u$ and $\theta$ was changed by $u = R\theta$, then the equation is ill posed. This suggests that the divergence free nature of $u$ must play an important role, unlike the critical and subcritical cases $s \geq 1/2$.


== Absolutely continuous measure ==
== Regularity of [[nonlocal minimal surfaces]] ==


In most cases, the nonnegative measure $\mu$ is assumed to be absolutely continuous: $\mathrm{d} \mu_x(y) = K(x,y) \mathrm{d}y$.
A nonlocal minimal surface that is sufficiently flat is known to be smooth. The possibility of singularities in the general case reduces to the analysis of a possible existence of nonlocal minimal cones. The problem can be stated as follows.


We keep this assumption in all the examples below.
For any $s \in (0,1)$, and any natural number $n$, is there any set $A \in \R^n$, other than a half space, such that
# $A$ is a cone: $\lambda A = A$ for any $\lambda > 0$.
# If $B$ is any set in $\R^n$ which coincides with $A$ outside of a compact set $C$, then the following inequality holds
\[ \int_C \int_{C} \frac{|\chi_A(x) - \chi_A(y)|}{|x-y|^{n+s}} \mathrm d x \mathrm d y + 2 \int_C \int_{\R^n \setminus C} \frac{|\chi_A(x) - \chi_A(y)|}{|x-y|^{n+s}} \mathrm d x \mathrm d y \leq \int_C \int_{C} \frac{|\chi_B(x) - \chi_B(y)|}{|x-y|^{n+s}} \mathrm d x \mathrm d y + 2\int_C \int_{\R^n \setminus C} \frac{|\chi_B(x) - \chi_B(y)|}{|x-y|^{n+s}} \mathrm d x \mathrm d y. \]


== Purely integro-differential operator ==
When $s$ is sufficiently close to one, such set does not exist if $n < 8$.


In this case we neglect the local part of the operator
== An integral ABP estimate ==
\[ Lu(x) = \int_{\R^n} (u(x+y) - u(x) - y \cdot \nabla u(x) \chi_{B_1}(y)) \, K(x,y) \mathrm d y. \]


== Symmetric kernels ==
The nonlocal version of the [[Alexadroff-Bakelman-Pucci estimate]] holds either for a right hand side in $L^\infty$ (in which the integral right hand side is approximated by a discrete sum) or under very restrictive assumptions on the kernels. Would the following result be true?
If the kernel is symmetric $K(x,y) = K(x,-y)$, then we can remove the gradient term from the integral and replace the difference by a second order quotient.


In the purely integro-differentiable case, it reads as
Assume $u_n \leq 0$ outside $B_1$ and for all $x \in B_1$,
\[ Lu(x) = \frac 12 \int_{\R^n} (u(x+y)+u(x+y)-2u(x)) \, K(x,y) \mathrm d y. \]
\[ \int_{\R^n} (u(x+y)-u(x)) K(x,y) \mathrm d y \geq \chi_{A_n}(x). \]
Where $\chi_{A_n}$ stands for the characteristic function of the sets $A_n$. Assume that the kernels $K$ satisfy symmetry and a uniform ellipticity condition
\begin{align*}
K(x,y) &= K(x,-y) \\
\lambda |y|^{-n-s} \leq K(x,y) &\leq \Lambda |y|^{-n-s} \qquad \text{for some } 0<\lambda<\Lambda \text{ and } s \in (0,2).
\end{align*}
If $|A_n|\to 0$ as $n \to +\infty$, is it true that $\sup u_n^+ \to 0$ as well?


The second order incremental quotient is sometimes abbreviated by $\delta u(x,y) := (u(x+y)+u(x+y)-2u(x))$.
== A local [[differentiability estimates|$C^{1,\alpha}$ estimate]] for integro-differential equations with nonsmooth kernels ==


== Translation invariant operators ==
Assume that $u : \R^n \to \R$ is a bounded function satisfying a [[fully nonlinear integro-differential equation]] $Iu=0$ in $B_1$. Assume that $I$ is elliptic with respect to the family of kernels $K$ such that
In this case, all coefficients are independent of $x$.
\[ \frac{\lambda(2-s)}{|y|^{n+s}} \leq K(y) \leq \frac{\Lambda(2-s)}{|y|^{n+s}}. \]
\[ Lu(x) = \mathrm{tr} \, A \cdot D^2 u + b \cdot \nabla u + c u + d + \int_{\R^n} (u(x+y) - u(x) - y \cdot \nabla u(x) \chi_{B_1}(y)) \, K(y) \mathrm{d}y. \]
Is it true that $u \in C^{1,\alpha}(B_1)$?


== The fractional Laplacian ==
An extra symmetry assumptions on the kernels may or maynot be necessary. The difficulty here is the lack of any smoothness assumption on the tails of the kernels $K$. This assumption is used in a localization argument in the proof of the [[differentiability estimates|$C^{1,\alpha}$ estimates]]. It is conceivable that the assumption may not be necessary at least for $s>1$.


The [[fractional Laplacian]] is the simplest and most common purely integro-differential operator. It corresponds to a translation invariant operator for which $K(y)$ is radially symmetric and homogeneous.
The need of the smoothness assumption for the $C^{1,\alpha}$ estimate is a subtle technical requirement. It is easy to overlook going through the proof naively.


\[ -(-\Delta)^{s/2} u(x) = C_{n,s} \int_{\R^n} (u(x+y)+u(x+y)-2u(x)) \frac{1}{|y|^{n+s}} \mathrm d y. \]
Note that the assumption is used only to localize an iteration of the [[Holder estimates]]. An equation of the form $Iu = f$ in the whole space $\R^n$ with $f \in C^\alpha$ would easily have $C^{1,\alpha}$ estimates without any smoothness restriction of the tails of the kernel.


== Uniformly elliptic of order $s$ ==
It is not clear how important or difficult this problem is. The solution may end up being a relatively simple technical approximation technique or may require a fundamentally new idea.


This corresponds to the assumption that the kernel is comparable to the one of the fractional Laplacian of the same order.
The same difficulty arises for $C^{s+\alpha}$ [[nonlocal Evans-Krylov theorem|estimates for convex equations]]. For example, is it true that a bounded function $u$ such that $M^+u = 0$ in $B_1$, where $M^+$ is the [[extremal operators|monster Pucci operator]] is $C^{s+\alpha}$ for some $\alpha>0$?
\[ \frac {(2-s)\lambda}{|y|^{n+s}} \leq K(x,y) \leq \frac {(2-s)\Lambda}{|y|^{n+s}}. \]


The normalizing factor $(2-s)$ is a normalizing factor which is only important when $s$ approaches two.
== A nonlocal generalization of the parabolic [[Krylov-Safonov theorem]] ==


An operator of variable order can be either one for which $s$ depends on $x$, or one for which there are two values $s_1<s_2$, one for the left hand side and another for the right hand side.
Let $u$ be a bounded function in $\R^n \times [-1,0]$ such that it solves an integro-differential parabolic equation
\[ u_t - \int_{\R^n} (u(x+y)-u(x)) K(x,y) \mathrm d y = 0 \qquad \text{in } B_1 \times (-1,0).\]
Making the usual symmetry and uniform ellipticity assumptions on the kernel $K$:
\begin{align*}
K(x,y) &= K(x,-y) \\
\frac{\lambda(2-s)}{ |y|^{-n-s}} \leq K(x,y) &\leq \frac{\Lambda(2-s)}{ |y|^{-n-s}} \qquad \text{for some } 0<\lambda<\Lambda \text{ and } s \in (0,2).
\end{align*}
Is it true that the solutions $u$ is Holder continuous in $B_{1/2} \times [-1/2,0]$, with an estimate
\[ ||u||_{C^\alpha(B_{1/2} \times [-1/2,0])} \leq C ||u||_{L^\infty(\R^n \times [-1,0])}, \]
for constants $C$ and $\alpha>0$ which do not blow up as $s \to 2$?


== Smoothness class $k$ of order $s$ ==
For an estimate with constants that blow up as $s \to 2$, one can easily adapt an argument for [[drift-diffusion equations]] <ref name="S2"/>.
This class (sometimes denoted as $\mathcal L_k^s$) corresponds to kernels that are uniformly elliptic of order $s$ and, moreover, their derivatives are also bounded
\[ |\partial_y^r K(x,y)| \leq \frac {\Lambda}{|y|^{n+s+r}} \ \ \text{for all } r\leq k. \]


== Order strictly below one ==
The elliptic version of this result is well known <ref name="CS"/>. The proof is not easy to adapt to the parabolic case because the [[Alexadroff-Bakelman-Pucci estimate]] is quite different in the elliptic and parabolic case.


If a non symmetric kernel $K$ satisfies the extra local integrability assumption
For gradient flows of Dirichlet forms, the problems appears open as well. However, it is conceivable that one could adapt the proof of the stationary case <ref name="K"/> to obtain the result without a major difficulty.  
\[ \int_{\R^n} \min(|y|,1) K(x,y) \mathrm d y < +\infty, \]
then the extra gradient term is not necessary in order to define the operator.
 
\[ Lu(x) = \int_{\R^n} (u(x+y) - u(x)) \, K(x,y) \mathrm d y. \]
 
The modification in the integro-differential part of the operator becomes an extra drift term.
 
A uniformly elliptic operator of order $s<1$ satisfies this condition.
 
== Order strictly above one ==
 
If a non symmetric kernel $K$ satisfies the extra integrability assumption on its tail.
\[ \int_{\R^n} \min(|y|^2,|y|) K(x,y) \mathrm d y < +\infty, \]
then the gradient term in the integral can be taken global instead of being cut off in the unit ball.
 
\[ Lu(x) = \int_{\R^n} (u(x+y) - u(x) - y \cdot \nabla u(x)) \, K(x,y) \mathrm d y. \]
 
The modification in the integro-differential part of the operator becomes an extra drift term.
 
A uniformly elliptic operator of order $s>1$ satisfies this condition.
 
== Indexed by a matrix ==
In some cases, it is interesting to study a family of kernels $K$ that are indexed by a matrix. For example, given the matrix $A$, one can consider the kernel of order $s$:
\[ K_A(y) =  \frac{(2-s) \langle y , Ay \rangle}{|y|^{n+2+s}}. \]
This family of kernels has the outstanding property that the corresponding linear operator $L$ coincides with $Lu(x) = a_{ij} \partial_{ij}\left[(-\Delta)^{(s-2)/2} u \right] (x)$ for some coefficients $a_{ij}$.
 
== Second order elliptic operators as limits of purely integro-differential ones ==
 
Given any bounded, even, positive function $a: \mathbb{R}^n\to \mathbb{R}$, the family of operators
 
\[L_\sigma u(x) = (2-\sigma) \int_{\mathbb{R}^n} (u(x+y)+u(x-y)-2u(x))\frac{a(y)}{|y|^{n+\sigma}}dy,\;\; \sigma \in (0,2), \]
 
define in the limit $\sigma \to 2^-$ a second order linear elliptic operator (possibly degenerate). This can be checked for any fixed $C^2$ function $u$ by a straightforward computation using the second order Taylor expansion. A class of kernels that is big enough to recover all translation invariant elliptic operators of the form  $Lu(x) = Tr ( A \cdot D^2u(x) )$ is given by the kernels
 
\[ K_A(y) = (2-\sigma) \frac{1}{|Ay|^{n+\sigma}},\]
 
where $A$ is an invertible symmetric matrix.
 
== Characterization via global maximum principle ==
 
A bounded linear operator
 
\[ L: C^2_0(\mathbb{R}^n) \to C(\mathbb{R}^n) \]
 
is said to satisfy the global maximum principle if given any $u \in C^2_0(\mathbb{R}^n)$ with a global maximum at some point $x_0$ we have
 
\[ (Lu)(x_0) \leq 0 \]
 
It turns out this property imposes strong restrictions on the operator $L$, and we have the following theorem due to Courrège <ref name="C65"/> <ref name="C64"/>: if $L$ satisfies the global maximum principle then it has the form
 
\[ Lu(x) = \mathrm{tr} \, A(x) \cdot D^2 u + b(x) \cdot \nabla u + c(x) u + \int_{\R^n} (u(x+y) - u(x) - y \cdot \nabla u(x) \chi_{B_1}(y)) \, \mathrm{d} \mu_x(y) \]
 
where again $A(x)$ is a nonnegative matrix for all $x$, $c(x)\leq 0$ and $\mu_x$ is a nonnegative measure for all $x$ satisfying
 
\[ \int_{\R^n} \min(y^2 , 1) \mathrm{d} \mu_x(y) < +\infty. \]
 
and $A(x),c(x)$ and $b(x)$ are bounded.
 
 
== See also ==
 
* [[Fractional Laplacian]]
* [[Levy processes]]
* [[Dirichlet form]]




== References ==
== References ==
{{reflist|refs=
{{reflist|refs=
<ref name="C64">{{Citation | last1=Courrège | first1=Philippe | title=Générateur infinitésimal d'un semi-groupe de convolution sur $R^n$, et formule de Lévy-Khinchine | year=1964 | journal=Bulletin des Sciences Mathématiques. 2e Série | issn=0007-4497 | volume=88 | pages=3–30}}</ref>
<ref name="CS">{{Citation | last1=Caffarelli | first1=Luis | last2=Silvestre | first2=Luis | title=Regularity theory for fully nonlinear integro-differential equations | url=http://dx.doi.org/10.1002/cpa.20274 | doi=10.1002/cpa.20274 | year=2009 | journal=[[Communications on Pure and Applied Mathematics]] | issn=0010-3640 | volume=62 | issue=5 | pages=597–638}}</ref>
<ref name="C65">{{Citation | last1=Courrège | first1=P. | title=Sur la forme intégro-différentielle des opéateurs de  $C_k^\infty(\mathbb{R}^n)$  dans $C(\mathbb{R}^n)$ satisfaisant au principe du maximum | journal=Sém. Théorie du potentiel (1965/66) Exposé | volume=2}}</ref>
<ref name="S2">{{Citation | last1=Silvestre | first1=Luis | title=Holder estimates for advection fractional-diffusion equations | year=To appear | journal=Annali della Scuola Normale Superiore di Pisa. Classe di Scienze}}</ref>
<ref name="K">{{Citation | last1=Kassmann | first1=Moritz | title=A priori estimates for integro-differential operators with measurable kernels | url=http://dx.doi.org/10.1007/s00526-008-0173-6 | doi=10.1007/s00526-008-0173-6 | year=2009 | journal=Calculus of Variations and Partial Differential Equations | issn=0944-2669 | volume=34 | issue=1 | pages=1–21}}</ref>
}}
}}

Revision as of 23:44, 6 June 2011

Well posedness of the supercritical surface quasi-geostrophic equation

Let $\theta_0 : \R^2 \to \R$ be a smooth function either with compact support or periodic. Let $s \in (0,1/2)$. Is there a global classical solution $\theta :\R^2 \to \R$ for the SQG equation? \begin{align*} \theta(x,0) &= \theta_0(x) \\ \theta_t + u \cdot \nabla \theta &= 0 \qquad \text{in } \R^2 \times (0,+\infty) \end{align*} where $u = R^\perp \theta$ and $R$ stands for the Riesz transform.

This is a very difficult open problem. It is believed that a solution would be a major step towards the understanding of Navier-Stokes equation. In the supercritical regime $s\in (0,1/2)$, the effect if the drift term is larger than the diffusion in small scales. Therefore, it seems unlikely that a proof of well posedness could be achieved with the methods currently known and listed in this wiki.

Note that if the relation between $u$ and $\theta$ was changed by $u = R\theta$, then the equation is ill posed. This suggests that the divergence free nature of $u$ must play an important role, unlike the critical and subcritical cases $s \geq 1/2$.

Regularity of nonlocal minimal surfaces

A nonlocal minimal surface that is sufficiently flat is known to be smooth. The possibility of singularities in the general case reduces to the analysis of a possible existence of nonlocal minimal cones. The problem can be stated as follows.

For any $s \in (0,1)$, and any natural number $n$, is there any set $A \in \R^n$, other than a half space, such that

  1. $A$ is a cone: $\lambda A = A$ for any $\lambda > 0$.
  2. If $B$ is any set in $\R^n$ which coincides with $A$ outside of a compact set $C$, then the following inequality holds

\[ \int_C \int_{C} \frac{|\chi_A(x) - \chi_A(y)|}{|x-y|^{n+s}} \mathrm d x \mathrm d y + 2 \int_C \int_{\R^n \setminus C} \frac{|\chi_A(x) - \chi_A(y)|}{|x-y|^{n+s}} \mathrm d x \mathrm d y \leq \int_C \int_{C} \frac{|\chi_B(x) - \chi_B(y)|}{|x-y|^{n+s}} \mathrm d x \mathrm d y + 2\int_C \int_{\R^n \setminus C} \frac{|\chi_B(x) - \chi_B(y)|}{|x-y|^{n+s}} \mathrm d x \mathrm d y. \]

When $s$ is sufficiently close to one, such set does not exist if $n < 8$.

An integral ABP estimate

The nonlocal version of the Alexadroff-Bakelman-Pucci estimate holds either for a right hand side in $L^\infty$ (in which the integral right hand side is approximated by a discrete sum) or under very restrictive assumptions on the kernels. Would the following result be true?

Assume $u_n \leq 0$ outside $B_1$ and for all $x \in B_1$, \[ \int_{\R^n} (u(x+y)-u(x)) K(x,y) \mathrm d y \geq \chi_{A_n}(x). \] Where $\chi_{A_n}$ stands for the characteristic function of the sets $A_n$. Assume that the kernels $K$ satisfy symmetry and a uniform ellipticity condition \begin{align*} K(x,y) &= K(x,-y) \\ \lambda |y|^{-n-s} \leq K(x,y) &\leq \Lambda |y|^{-n-s} \qquad \text{for some } 0<\lambda<\Lambda \text{ and } s \in (0,2). \end{align*} If $|A_n|\to 0$ as $n \to +\infty$, is it true that $\sup u_n^+ \to 0$ as well?

A local $C^{1,\alpha}$ estimate for integro-differential equations with nonsmooth kernels

Assume that $u : \R^n \to \R$ is a bounded function satisfying a fully nonlinear integro-differential equation $Iu=0$ in $B_1$. Assume that $I$ is elliptic with respect to the family of kernels $K$ such that \[ \frac{\lambda(2-s)}{|y|^{n+s}} \leq K(y) \leq \frac{\Lambda(2-s)}{|y|^{n+s}}. \] Is it true that $u \in C^{1,\alpha}(B_1)$?

An extra symmetry assumptions on the kernels may or maynot be necessary. The difficulty here is the lack of any smoothness assumption on the tails of the kernels $K$. This assumption is used in a localization argument in the proof of the $C^{1,\alpha}$ estimates. It is conceivable that the assumption may not be necessary at least for $s>1$.

The need of the smoothness assumption for the $C^{1,\alpha}$ estimate is a subtle technical requirement. It is easy to overlook going through the proof naively.

Note that the assumption is used only to localize an iteration of the Holder estimates. An equation of the form $Iu = f$ in the whole space $\R^n$ with $f \in C^\alpha$ would easily have $C^{1,\alpha}$ estimates without any smoothness restriction of the tails of the kernel.

It is not clear how important or difficult this problem is. The solution may end up being a relatively simple technical approximation technique or may require a fundamentally new idea.

The same difficulty arises for $C^{s+\alpha}$ estimates for convex equations. For example, is it true that a bounded function $u$ such that $M^+u = 0$ in $B_1$, where $M^+$ is the monster Pucci operator is $C^{s+\alpha}$ for some $\alpha>0$?

A nonlocal generalization of the parabolic Krylov-Safonov theorem

Let $u$ be a bounded function in $\R^n \times [-1,0]$ such that it solves an integro-differential parabolic equation \[ u_t - \int_{\R^n} (u(x+y)-u(x)) K(x,y) \mathrm d y = 0 \qquad \text{in } B_1 \times (-1,0).\] Making the usual symmetry and uniform ellipticity assumptions on the kernel $K$: \begin{align*} K(x,y) &= K(x,-y) \\ \frac{\lambda(2-s)}{ |y|^{-n-s}} \leq K(x,y) &\leq \frac{\Lambda(2-s)}{ |y|^{-n-s}} \qquad \text{for some } 0<\lambda<\Lambda \text{ and } s \in (0,2). \end{align*} Is it true that the solutions $u$ is Holder continuous in $B_{1/2} \times [-1/2,0]$, with an estimate \[ ||u||_{C^\alpha(B_{1/2} \times [-1/2,0])} \leq C ||u||_{L^\infty(\R^n \times [-1,0])}, \] for constants $C$ and $\alpha>0$ which do not blow up as $s \to 2$?

For an estimate with constants that blow up as $s \to 2$, one can easily adapt an argument for drift-diffusion equations [1].

The elliptic version of this result is well known [2]. The proof is not easy to adapt to the parabolic case because the Alexadroff-Bakelman-Pucci estimate is quite different in the elliptic and parabolic case.

For gradient flows of Dirichlet forms, the problems appears open as well. However, it is conceivable that one could adapt the proof of the stationary case [3] to obtain the result without a major difficulty.


References

  1. Silvestre, Luis (To appear), "Holder estimates for advection fractional-diffusion equations", Annali della Scuola Normale Superiore di Pisa. Classe di Scienze 
  2. Caffarelli, Luis; Silvestre, Luis (2009), "Regularity theory for fully nonlinear integro-differential equations", Communications on Pure and Applied Mathematics 62 (5): 597–638, doi:10.1002/cpa.20274, ISSN 0010-3640, http://dx.doi.org/10.1002/cpa.20274 
  3. Kassmann, Moritz (2009), "A priori estimates for integro-differential operators with measurable kernels", Calculus of Variations and Partial Differential Equations 34 (1): 1–21, doi:10.1007/s00526-008-0173-6, ISSN 0944-2669, http://dx.doi.org/10.1007/s00526-008-0173-6