next up previous contents
Next: Maximum Principles Up: PARTIAL DIFFERENTIAL EQUATIONS (PDE's) Previous: Splitting and ADI. Nonlinear   Contents

ELLIPTIC EQUATIONS

Brief overview Refer to Figure 30 Archetypical Equation in 2D:

$\displaystyle \nabla^2 U=f(x,y)$    in a domain $\displaystyle \Omega(x,y).
$

with $\nabla^2\equiv \partial _{xx}+\partial _{yy}$

is called Poisson's Equation. If $f=0$ we call it Laplace's equation. The solutions to Laplace's equations are called harmonic functions and are intimately tied to the theory of complex analysis.

Figure: Domain of definition for Poisson's Problem in 2 dimensions. The $\hat n$ indicates the convention on the unit normal to the boundary $\partial \Omega .$
\includegraphics[height=3in]{dom.eps}

Boundary Conditions:

(1)
(field specified at boundary) Dirchlet: $U=b_1$ on $\partial \Omega$
(2)
Neumann: $\displaystyle \frac{\partial U}{\partial n}\equiv\hat n\cdot \nabla
U=b_2$ on $\partial \Omega$
(3)
Robin: mix of the two above: $\displaystyle \frac{\partial U}{\partial n}+\alpha
U=b_3$ on $\partial \Omega$. Only 1 boundary condition may be specified on $\partial \Omega$. But can have Neumann, say, for one portion of $\partial \Omega$ and Dirichlet for the other, for example.
Only 1 boundary condition may be specified on $\partial \Omega$. But can have Neumann, say, for one portion of $\partial \Omega$ and Dirichlet for the other, for example.

Physically: Poisson equation describes many things. For example, the steady state-temperatue distribution of an object occupying $\Omega $, with heat sources and sinks represented by $f$. The Dirichlet boundary conditions represent the situation when the temperature is specified at boundary and Naumann would be if the flux of temperature is specified at boundary. In particular, if $\displaystyle \frac{\partial
U}{\partial n}=0\vert _{\partial \Omega}$ we say we have a perfect insulator boundary.

In order for solution to exist, if Newmann B.C. are specified, is the data must satisfy the ``integrability'' condition:

$\displaystyle \int\int_{\Omega}fdV=\int_{\partial \Omega}b_2ds
$

(to prove: use divergence theorem).

General Quasi-linear 2nd-order elliptic in 2D has the form:

$\displaystyle a(x,y)U_{xx}+2b(x,y)U_{xy}+c(x,y)U_{yy}+d(x,y,U,U_x,U_y)=f(x,y)
$

with $a,c>0$     $b^2<ac$

A $1^{st}$-order elliptic equation system example:

$\displaystyle \left\{\begin{array}{l}
U_x-V_y=0\qquad \mbox {\lq\lq Cauchy-Riemann'' equations}\\
U_y+V_x=0
\end{array}\right.
$

example of $4^{th}$ order

$\displaystyle \nabla^4 U=f$$\displaystyle \qquad \mbox {\lq\lq Biharmonic Equation.''}
$

An essential feature of elliptic equation solutions is that they are smoother than the data. For example $U$ has 2 more derivatives then $f$ in the Poisson equation. 4 more than $f$ in biharmonic equation. Solutions to Laplace and Cauchy-Riemann Eqs are infinitely differentiable.

For the general $2^{nd}$-order linear constant coefficient elliptic equations, the following ``regularity estimate'' can be proved:

$\displaystyle \vert\vert U\vert\vert^2_{s+2}\le C_s\left(\vert\vert f\vert\vert^2_s+\vert\vert U\vert\vert^2_0\right)
$

where

$\displaystyle \vert\vert\cdot\vert\vert^2_s\equiv \sum_{s_1+s_2\le s}\vert\vert\partial ^{s_{1}}_x\partial ^{s_2}_y
\cdot \vert\vert^2
$

i.e. if solution exists and is finite in $L_2$, i.e. $\vert\vert U\vert\vert _0$ finite and that $f$ has all derivatives of order up to $s$ in $L_2({\Bbb R}^2)\Rightarrow U$ has $s+2$ in $L_2(R_2)$.

The solution of the elliptic equation is more differentiable than the data and the increase in differentiability $=$ order of equation.

There is an ``interior regularity estimate'' as well. Suppose $\Omega_1  \in   \Omega$ whose boundary is wholly contained in $\Omega $. Then

$\displaystyle \vert\vert U\vert\vert^2_{s+2, \Omega_1}\le C_s(\Omega,
\Omega_1)(\vert\vert f\vert\vert^2_{s,\Omega}+\vert\vert U\vert\vert^2_{0, \Omega})
$

For the non-constant coefficient case, we require that coefficients be defined and bounded, and very similar estimates are obtained.



Subsections
next up previous contents
Next: Maximum Principles Up: PARTIAL DIFFERENTIAL EQUATIONS (PDE's) Previous: Splitting and ADI. Nonlinear   Contents
Juan Restrepo 2003-05-02