2
$\begingroup$

Consider a First-order PDE in $2D$, i.e. find a function $u : \Omega \subset \mathbb{R}^2 \rightarrow \mathbb{R}$ such that $u_x = p$ and $u_y = q$ for some functions $p, q: \Omega \rightarrow \mathbb{R}$

What kind of numerical scheme might one use to solve this PDE, if we only have grid functions $p^h, q^h$ on some discretization $\Omega^h$ of $\Omega$? What kind of boundary conditions can we apply?

Although this setup seems to me like the easiest multidimensional PDE, most scripts about PDE or numerics for PDE in $2D$ start with some $5$-point star for $\Delta u = 0$, a second order PDE. I have not been able to find a simple go-to scheme to solve first order PDEs.

$\endgroup$
1
$\begingroup$

Here is a simple approach, for the case where $\Omega$ is a rectangle. There is only a solution if $p_y = q_x$. The value of $u$ must be specified at some specific point $(x_0,y_0)$. The value of $u$ at any other point can be computed by integrating: $u(x,y) = u(x_0,y_0) + \int_{x_0}^x u_x(s,y_0) ds + \int_{y_0}^y u_y(x,t) dt$. These integrals can be approximated numerically.

$\endgroup$
  • $\begingroup$ Thank you for the good answer, but it is not exactly what I was looking for. I am interested in concrete schemes to solve the problem. I'd love to see something like convergence results. Although I think I'd be able to derive a scheme from what you wrote, there must be some beginners course where those results are already proved. $\endgroup$ – don-joe Feb 26 '18 at 12:48
1
$\begingroup$

I'm kind of thinking you already have all the pieces of the puzzle you need and just have to assemble them into a solution. You have $$\vec\nabla u(x,y)=p(x,y)\hat i+q(x,y)\hat j$$ So first off, as littleO explained $$\vec\nabla\times\vec\nabla u=\vec0=\left(\frac{\partial q}{\partial x}-\frac{\partial p}{\partial y}\right)\hat k$$ Assuming your data $p(x,y)$ and $q(x,y)$ pass this test we can now evaluate $$\vec\nabla\cdot\vec\nabla u=\nabla^2u=\frac{\partial p}{\partial x}+\frac{\partial q}{\partial y}$$ You know that $$\begin{align}\nabla^2u(x,y)&=\frac{u(x+h,y)+u(x,y+h)+u(x-h,y)+u(x,y-h)-4u(x,y)}{h^2}+O(h^2)\\ \frac{\partial p}{\partial x}&=\frac{p(x+h)-p(x-h)}{2h}+O(h^2)\\ \frac{\partial q}{\partial y}&=\frac{q(y+h)-q(y-h)}{2h}+O(h^2)\end{align}$$ And we are given Neumann boundary conditions: $$\hat n\cdot\vec\nabla u(x,y)=\hat n\cdot(p(x,y)\hat i+q(x,y)\hat j)$$ Which I am assuming you already know how to discretize.

$\endgroup$
  • $\begingroup$ So we basically transform the PDE into a Poisson problem? Is that a typical approach to tackle this class of PDEs? $\endgroup$ – don-joe Feb 27 '18 at 15:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.