List of results that are fundamentally different to the local case and De Giorgi-Nash-Moser theorem: Difference between pages

From nonlocal pde
(Difference between pages)
Jump to navigation Jump to search
imported>Luis
No edit summary
 
imported>Luis
 
Line 1: Line 1:
In this page we collect some results in nonlocal equations that contradict the intuition built in analogy with the local case.
The De Giorgi-Nash-Moser theorem provides Holder estimates and the Harnack inequality for uniformly elliptic or parabolic equations with rough coefficients in divergence form.


The list is very incomplete right now. Please help expand it by editing it.
The equation is
\[ \mathrm{div} A(x) \nabla u(x) = \partial_i a_{ij}(x) \partial_j u(x) = 0, \]
in the elliptic case, or
\[ u_t = \mathrm{div} A(x,t) \nabla u(x). \]
Here $A = \{a_{ij}\}$ is a matrix valued function in $L^\infty$ satisfying the uniform ellipticity condition for some $\lambda>0$,
\[ \langle A v,v \rangle \geq \lambda |v|^2,\]
for every $v \in \R^n$, uniformly in space and time.


=== Traveling fronts in Fisher-KPP equations with fractional diffusion have exponential speed ===
The corresponding result in non divergence form is [[Krylov-Safonov theorem]].
Let us consider the  reaction diffusion equation
\[ u_t + (-\Delta)^s u = f(u), \]
with a Fisher-KPP type of nonlinearity (for example $f(u) = u(1-u)$). In the local diffusion case, the stable state $u=1$ invades the unstable state $u=0$ at a constant speed. In the nonlocal case (any $s<1$), the invasion holds at an exponential rate.


The explanation of the difference can be understood intuitively from the fact that the fat tails in the fractional heat kernels make diffusion happen at a much faster rate <ref name="CR"/><ref name="CR2"/>.
For nonlocal equations, there are analogous results both for [[Holder estimates]] and the [[Harnack inequality]].


=== The optimal regularity for the fractional obstacle problem exceeds the scaling of the equation ===
== Elliptic version ==
Given a function $\varphi$, the obstacle problem consists in the solution to an equation of the form
For the result in the elliptic case, we assume that the equation
\[ \min((-\Delta)^s u , u-\varphi) = 0.\]
\[ \mathrm{div} A(x) \nabla u(x) = 0 \]
is satisfied in the unit ball $B_1$ of $\R^n$.
===Holder estimate===
The Holder estimate says that if $u$ is an $L^2$ solution to a uniformly elliptic divergence form equation as above, then $u$ is Holder continuous in $B_{1/2}$ and
\[ ||u||_{C^\alpha(B_{1/2})} \leq C ||u||_{L^2(B_1)}.\]
The constants $C$ and $\alpha>0$ depend on $n$ (dimension), $\lambda$ and $||A||_{L^\infty}$.


If $\varphi$ is smooth enough, the solution $u$ to the obstacle problem will be $C^{1,s}$ and no better. There is a big difference between the case $s=1$ and $s<1$ which makes the proof fundamentally different. In the classical case $s=1$, the optimal regularity matches the scaling of the equation. The classical proof of optimal regularity is to show an upper bound in the separation of $u$ from the obstacle in the unit ball and then just scale it. In the fractional case $s<1$, this method only gives $C^{2s}$ regularity, which matches the scaling of the equation. It is somewhat surprising that a better regularity result holds and it requires a different method for the proof.
The result can be scaled to balls of arbitrary radius $r>0$ to obtain
\[ [u]_{C^\alpha(B_{r/2})} \leq C \frac{||u||_{L^2(B_r)}}{r^\alpha}.\]


The intuitive explanation is that $(-\Delta)^s u$ satisfies an extra elliptic equation in terms of its Laplacian to the power $1-s$, and that equation provides the extra regularity <ref name="S"/>.
Moreover, by covering an arbitrary domain $\Omega$ with balls, one can show that a solution to the equation in $\Omega$ is $C^\alpha$ in the interior of $\Omega$.


=== Solutions to nonlocal elliptic equations can have interior maximums ===
===Harnack inequality===
The [[Harnack inequality]] says that if $u$ is a non negative solution of the equation in $B_1$, then its minimum controls its maximum in $B_{1/2}$:
\[ \max_{B_{1/2}} u \leq C \min_{B_{1/2}} u.\]
The constant $C$ depends on $n$, $\lambda$ and $||A||_{L^\infty}$ only.


Solutions to linear (and nonlinear) integro-differential equations satisfy a ''nonlocal'' maximum principle: they cannot have a ''global'' maximum or minumum in the interior of the domain of the equation. Local extrema are possible.
===Minimizers of convex functionals===
The theorem of De Giorgi, Nash and Moser was used originally to solve one of the famous Hilbert problems. The question was whether the minimizers of Dirichlet integrals
\[ J(u) := \int_{\Omega} F(\nabla u) \mathrm{d} x,\]
are always smooth if $F$ is smooth and strictly convex. The theorem of De Giorgi-Nash-Moser in its elliptic form can be applied to the differential quotients of the minimizer of $J$ to show that the solution is $C^{1,\alpha}$. Once that initial regularity is obtained, further regularity follows by [[bootstrapping]] with the [[Schauder estimates]] and the smoothness of $F$.


This is related to the fact that Dirichlet boundary conditions have to be given in the whole complement of the Domain and not only on its boundary. It is also related to the failure in general of the classical [[Harnack inequality]] unless the positivity of the function is assumed in the full space <ref name="K"/>.
Note that in order to apply the theorem to these nonlinear equations, it is very important that no smoothness assumption on the coefficients $A(x)$ is made.


In fact, any function $f\in C^k(\overline{B_1})$ can by approximated with a solution to $(-\Delta)^su$ in $B_1$ that vanishes outside a compact set <ref name="dipierro2014all" />. That is, s-harmonic functions are dense in $C^k_{loc}$. This is clearly in contrast with the rigidity of harmonic functions, and is a purely nonlocal feature.
== Parabolic version ==
For the result in the parabolic case, we assume that the equation
\[ u_t - \mathrm{div} A(x) \nabla u(x) = 0 \]
is satisfied in the unit cylinder $(0,1] \times B_1$ of $\R \times \R^n$.
===Holder estimate===
The Holder estimate says that if $u$ is an $L^2$ solution to a uniformly elliptic divergence form equation as above, then $u$ is Holder continuous in $[1/2,1] \times B_{1/2}$ and
\[ ||u||_{C^\alpha([1/2,1] \times B_{1/2})} \leq C ||u||_{L^2([0,1] \times B_1)}.\]
The constants $C$ and $\alpha>0$ depend on $n$ (dimension), $\lambda$ and $||A||_{L^\infty}$.


=== Boundary regularity of solutions is different from the interior ===
===Harnack inequality===
The [[Harnack inequality]] says that if $u$ is a non negative solution of the equation in $[0,1] \times B_1$, then its minimum controls its maximum in a previous time:
\[ \sup_{[1/4,1/2] \times B_{1/2}} u \leq \inf_{[3/4,0] \times B_{1/2}} u. \]


For second order equations, the boundary regularity of solutions to $\Delta u=0$ is the same as in the interior. For example, a solution to $\Delta u=0$ in $B_1^+$, with $u=0$ in $\{x_n=0\}$, can be extended (by odd reflection) to a solution of $\Delta u=0$ in $B_1$. Thus, in this case the boundary regularity of $u$ just follows from the interior regularity --one has $u\in C^\infty(\overline{B_{1/2}^+})$.
===Gradient flows===
 
The parabolic version of the theory can be used to show that the solutions to gradient flow equations with strictly convex energies are smooth.
In a general smooth domain $\Omega$ one can flatten the boundary and repeat the previous argument to get that $u\in C^\infty(\overline\Omega)$.
\[ u_t + \partial_u J[u] = u_t - \mathrm{div} \left( (\partial_i F)(\nabla u) \partial_i u \right) = 0.\]
 
The idea of the proof is that the derivatives of $u$ (or its differential quotients) satisfy an equation with rough but uniformly elliptic coefficients.
This is not the case for nonlocal equations. Indeed, the function $u(x)=(x_+)^s$ satisfies $(-\Delta)^su=0$ in $(0,\infty)$, and $u=0$ in $(-\infty,0)$. However, $u$ is not even Lipschitz up to the boundary, while all solutions are $C^\infty$ in the interior. This is related to the fact that the odd reflection of $u$ is not anymore a solution to the same equation.
 
More generally, solutions to $(-\Delta)^su=f$ in $\Omega$, with $u=0$ in $\mathbb R^n\setminus\Omega$, are smooth in the interior of $\Omega$, but not up to the boundary. The optimal Holder regularity is $u\in C^s(\overline\Omega)$. See [[boundary regularity for integro-differential equations]] for more details.
 
=== For some equations, the weak Harnack inequality may hold while the full [[Harnack inequality]] does not ===
 
The weak Harnack inequality relates the minimum of a positive supersolution to an elliptic equation to its $L^p$ norm. It is an important step used to derive [[Hölder estimates]] and also the usual [[Harnack inequality]]. However, there are examples of non local elliptic equation for which the weak Harnack inequality and Hölder estimates hold, whereas the classical Harnack inequality does not. There is a discussion about this fact in an article by Moritz Kassmann, Marcus Rang and Russell Schwab <ref name="rang2013h" />.
 
=== Solutions to elliptic linear and translation invariant equations may not be smooth ===
 
For second order equations, any solution to an elliptic linear and translation invariant equation $Lu=f$ in $\Omega$ is smooth in the interior whenever $f$ is smooth. For second order equations, $L$ must be of the form $Lu=a_{ij}\partial_{ij}u$, and hence after an affine change of variables it is just the Laplacian $\Delta$.
 
For nonlocal equations, solutions to $Lu=f$ in $\Omega$, with $f$ smooth, may not be smooth inside $\Omega$, even if $L$ is an elliptic linear and translation invariant operator like
\[Lu(x)=\int_{\mathbb R^n}\bigl(u(x+y)-u(x)\bigr)K(y)dy,\]
with $K(y)=K(-y)$ and satisfying
\[\frac{\lambda}{|y|^{n+2s}}\leq K(y)\leq \frac{\Lambda}{|y|^{n+2s}}.\]
It was proved in <ref name="Serra2" /> that there exist a solution to $Lu=0$ in $B_1$, with $u\in L^\infty(\R^n)$, which is not $C^{2s+\epsilon}(B_{1/2})$ for any $\epsilon>0$. The counterexample can be constructed even in dimension 1, and it is very related to the regularity of the kernel $K$.
 
Related to this, it was shown in <ref name="Ros-Valdinoci" /> that there is a $C^\infty$ domain $\Omega$ and an operator of the form
\[Lu(x)=\int_{\mathbb R^n}\bigl(u(x+y)-u(x)\bigr)\frac{a(y/|y|)}{|y|^{n+2s}}\,dy,\]
for which the solution to $Lu=1$ in $\Omega$, $u=0$ in $\mathbb R^n\setminus\Omega$, is not $C^{3s+\epsilon}$ inside $\Omega$ for any $\epsilon>0$. See the survey <ref name="Ros-survey" /> for more details.
 
 
=== Viscosity solutions can be evaluated at points ===
 
The concept of [[viscosity solutions]] is developed in order to make sense of an elliptic equation even for continuous functions for which the equation cannot be evaluated classically at points. The idea is to use test functions whose graphs are tangent to the graphs of the weak solution at some point, and then evaluate the equation on that test function. The ellipticity property tells us that the value of the equation for that test function at the point of contact must have certain sign, and this is the condition that a viscosity solution fulfill.
 
It turns out that for a large class of [[fully nonlinear integro-differential equations]], every time a viscosity solution can be touched by a smooth test function at a point, then the equation can be evaluated classically for '''the original function''' at that point <ref name="CS"/>.
 
=== Viscosity solutions to fully nonlinear integro-differential equations can be approximated with classical solutions ===
It is a very classical trick that if we have a weak solution to a linear PDE with constant coefficients, we can approximate it with a smooth solution via a simple mollification. For nonlinear equations this trick is no longer available and we are always forced to deal with the technical difficulties of viscosity solutions. This is an apparent difficulty for example when proving [[regularity estimates]], since in general we cannot derive them an a priori estimate for a classical solution. On the other hand, [[viscosity solutions]] to [[fully nonlinear integro-differential equations]] can be approximated by $C^2$ solutions to approximate equations <ref name="CS3"/>.
 
This procedure can also be carried out to approximate viscosity solutions to fully nonlinear elliptic equations with smooth solutions of an approximate equation. However, the approximated equation is integro-differential <ref name="CS4"/>.
 
=== Improved differentiability of solutions to integro-differential equations in divergence form ===
 
A classical theorem asserts that solution to uniformly elliptic equations in divergence form
\[ \mathrm{div} \, ( A(x) \nabla u) = 0,\]
where $\lambda I \leq A(x) \leq \Lambda I$ belong to the space $W^{1,2+\varepsilon}$ for some $\varepsilon > 0$. This is a nontrivial result, since the variational formulation of the problem only gives us a solution in $W^{1,2}$. The result provides an improvement in the integrability of $|\nabla u|$ from $L^2$ to $L^{2+\varepsilon}$.
 
== References ==
{{reflist|refs=
<ref name="CR">{{Citation | last1=Cabré | first1=Xavier | last2=Roquejoffre | first2=Jean-Michel | title=Propagation de fronts dans les équations de Fisher-KPP avec diffusion fractionnaire | url=http://dx.doi.org/10.1016/j.crma.2009.10.012 | doi=10.1016/j.crma.2009.10.012 | year=2009 | journal=Comptes Rendus Mathématique. Académie des Sciences. Paris | issn=1631-073X | volume=347 | issue=23 | pages=1361–1366}}</ref>
<ref name="CR2">{{Citation | last1=Cabré | first1=Xavier | last2=Roquejoffre | first2=Jean-Michel | title=The influence of fractional diffusion in Fisher-KPP equations | url=http://arxiv.org/abs/1202.6072 | year=to appear | journal=Comm. Math. Phys.}}</ref>
<ref name="S">{{Citation | last1=Silvestre | first1=Luis | title=Regularity of the obstacle problem for a fractional power of the Laplace operator | url=http://dx.doi.org/10.1002/cpa.20153 | doi=10.1002/cpa.20153 | year=2007 | journal=[[Communications on Pure and Applied Mathematics]] | issn=0010-3640 | volume=60 | issue=1 | pages=67–112}}</ref>
<ref name="K">{{Citation | last1=Kassmann | first1=Moritz | title=The classical Harnack inequality fails for non-local operators | year=Preprint}}</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="CS3">{{Citation | last1=Caffarelli | first1=Luis | last2=Silvestre | first2=Luis | title=The Evans-Krylov theorem for non local fully non linear equations | year=to appear | journal=[[Annals of Mathematics]] | issn=0003-486X}}</ref>
<ref name="CS4">{{Citation | last1=Caffarelli | first1=Luis | last2=Silvestre | first2=Luis | title=Nonlinear partial differential equations and related topics | publisher=[[American Mathematical Society]] | location=Providence, R.I. | series=Amer. Math. Soc. Transl. Ser. 2 | year=2010 | volume=229 | chapter=Smooth approximations of solutions to nonconvex fully nonlinear elliptic equations | pages=67–85}}</ref>
<ref name="rang2013h">{{Citation | last1=Rang | first1= Marcus | last2=Kassmann | first2= Moritz | last3=Schwab | first3= Russell W | title=H$\backslash$" older Regularity For Integro-Differential Equations With Nonlinear Directional Dependence | journal=arXiv preprint arXiv:1306.0082}}</ref>
<ref name="dipierro2014all">{{Citation | last1=Dipierro | first1= Serena | last2=Savin | first2= Ovidiu | last3=Valdinoci | first3= Enrico | title=All functions are locally $s$-harmonic up to a small error | journal=arXiv preprint arXiv:1404.3652}}</ref>
<ref name="Serra2">{{Citation | last1=Serra | first1= Joaquim | title=$C^{2s+\alpha}$ regularity for concave nonlocal fully nonlinear elliptic equations with rough kernels | journal=arXiv preprint}}</ref>
<ref name="Ros-Valdinoci">{{Citation | last1=Ros-Oton | first1= Xavier | last2=Valdinoci | first2= Enrico | title=The Dirichlet problem for nonlocal operators with singular kernels: convex and nonconvex domains | journal=arXiv preprint}}</ref>
<ref name="Ros-survey">{{Citation | last1=Ros-Oton | first1= Xavier | title=Nonlocal elliptic equations in bounded domains: a survey | journal=arXiv preprint}}</ref>
<ref name="kuusi2015">{{Citation | last1=Kuusi | first1= Tuomo | last2=Mingione | first2= Giuseppe | last3=Sire | first3= Yannick | title=Nonlocal self-improving properties | url=http://dx.doi.org/10.2140/apde.2015.8.57 | journal=Anal. PDE | issn=2157-5045 | year=2015 | volume=8 | pages=57--114 | doi=10.2140/apde.2015.8.57}}</ref>
}}

Revision as of 15:14, 14 March 2012

The De Giorgi-Nash-Moser theorem provides Holder estimates and the Harnack inequality for uniformly elliptic or parabolic equations with rough coefficients in divergence form.

The equation is \[ \mathrm{div} A(x) \nabla u(x) = \partial_i a_{ij}(x) \partial_j u(x) = 0, \] in the elliptic case, or \[ u_t = \mathrm{div} A(x,t) \nabla u(x). \] Here $A = \{a_{ij}\}$ is a matrix valued function in $L^\infty$ satisfying the uniform ellipticity condition for some $\lambda>0$, \[ \langle A v,v \rangle \geq \lambda |v|^2,\] for every $v \in \R^n$, uniformly in space and time.

The corresponding result in non divergence form is Krylov-Safonov theorem.

For nonlocal equations, there are analogous results both for Holder estimates and the Harnack inequality.

Elliptic version

For the result in the elliptic case, we assume that the equation \[ \mathrm{div} A(x) \nabla u(x) = 0 \] is satisfied in the unit ball $B_1$ of $\R^n$.

Holder estimate

The Holder estimate says that if $u$ is an $L^2$ solution to a uniformly elliptic divergence form equation as above, then $u$ is Holder continuous in $B_{1/2}$ and \[ ||u||_{C^\alpha(B_{1/2})} \leq C ||u||_{L^2(B_1)}.\] The constants $C$ and $\alpha>0$ depend on $n$ (dimension), $\lambda$ and $||A||_{L^\infty}$.

The result can be scaled to balls of arbitrary radius $r>0$ to obtain \[ [u]_{C^\alpha(B_{r/2})} \leq C \frac{||u||_{L^2(B_r)}}{r^\alpha}.\]

Moreover, by covering an arbitrary domain $\Omega$ with balls, one can show that a solution to the equation in $\Omega$ is $C^\alpha$ in the interior of $\Omega$.

Harnack inequality

The Harnack inequality says that if $u$ is a non negative solution of the equation in $B_1$, then its minimum controls its maximum in $B_{1/2}$: \[ \max_{B_{1/2}} u \leq C \min_{B_{1/2}} u.\] The constant $C$ depends on $n$, $\lambda$ and $||A||_{L^\infty}$ only.

Minimizers of convex functionals

The theorem of De Giorgi, Nash and Moser was used originally to solve one of the famous Hilbert problems. The question was whether the minimizers of Dirichlet integrals \[ J(u) := \int_{\Omega} F(\nabla u) \mathrm{d} x,\] are always smooth if $F$ is smooth and strictly convex. The theorem of De Giorgi-Nash-Moser in its elliptic form can be applied to the differential quotients of the minimizer of $J$ to show that the solution is $C^{1,\alpha}$. Once that initial regularity is obtained, further regularity follows by bootstrapping with the Schauder estimates and the smoothness of $F$.

Note that in order to apply the theorem to these nonlinear equations, it is very important that no smoothness assumption on the coefficients $A(x)$ is made.

Parabolic version

For the result in the parabolic case, we assume that the equation \[ u_t - \mathrm{div} A(x) \nabla u(x) = 0 \] is satisfied in the unit cylinder $(0,1] \times B_1$ of $\R \times \R^n$.

Holder estimate

The Holder estimate says that if $u$ is an $L^2$ solution to a uniformly elliptic divergence form equation as above, then $u$ is Holder continuous in $[1/2,1] \times B_{1/2}$ and \[ ||u||_{C^\alpha([1/2,1] \times B_{1/2})} \leq C ||u||_{L^2([0,1] \times B_1)}.\] The constants $C$ and $\alpha>0$ depend on $n$ (dimension), $\lambda$ and $||A||_{L^\infty}$.

Harnack inequality

The Harnack inequality says that if $u$ is a non negative solution of the equation in $[0,1] \times B_1$, then its minimum controls its maximum in a previous time: \[ \sup_{[1/4,1/2] \times B_{1/2}} u \leq \inf_{[3/4,0] \times B_{1/2}} u. \]

Gradient flows

The parabolic version of the theory can be used to show that the solutions to gradient flow equations with strictly convex energies are smooth. \[ u_t + \partial_u J[u] = u_t - \mathrm{div} \left( (\partial_i F)(\nabla u) \partial_i u \right) = 0.\] The idea of the proof is that the derivatives of $u$ (or its differential quotients) satisfy an equation with rough but uniformly elliptic coefficients.