Stepping backwards with Forward Euler?

225 Views Asked by At

Let us say I want to use Forward Euler scheme to solve the heat equation $$ \frac{\partial u}{\partial t} = -\frac{\partial^2 u}{\partial x^2} $$ in the domain $t \in (0, 1)$, $x \in (0,1)$ but instead of an initial condition $u(x, 0)$ I have the end condition $u(x, t=1)$. How will then the Forward Euler look? Can I simply step backwards like this: $$ \frac{u_j^{n-1} - u_j^n}{\Delta t} = - \left( \frac{u_{j+1}^{n} -2u_j^{n} + u_{j-1}^n}{(\Delta x)^2} \right) $$ where $n$ are the time-steps and $j$ are the $x$-steps. The reason I have doubts about this solution is that when I try to look at the truncation error for this scheme by Taylor-expanding LHS at $t_n$ and RHS at $x_j$, I end up with the expression $$ -\frac{\partial u}{\partial t} + O(\Delta t) = -\frac{\partial^2 u}{\partial x^2} + O(\Delta x) $$ which does not allow me to substitute the exact solution $\frac{\partial u}{\partial t} = -\frac{\partial^2 u}{\partial x^2}$ since the expressions have opposite signs. I don't see that my Taylor expansions have any errors, so it might seem that the whole method is wrong?

Any help much appreciated.

1

There are 1 best solutions below

1
On BEST ANSWER

Note that equation you're solving is not exactly the heat equation, since the heat equation is $$ \frac{\partial u}{\partial t} = \varkappa \frac{\partial^2 u}{\partial x^2}, \qquad \varkappa > 0, $$ but If you're solving your problem backward in time, the problem becomes well posed only when $\varkappa < 0$, and you need to be careful when approximating the time derivative with a finite difference. Whatever the time direction is the time derivative is approximated as $$ \frac{u^{n} - u^{n-1}}{\Delta t} $$ assuming $\Delta t > 0$ and as $$ \frac{u^{n-1} - u^{n}}{\Delta t} $$ if $\Delta t < 0$. That's a matter of convention, no more. Since you've moving backward in time the $\Delta t = t_{n-1} - t_n < 0$. Thus $$ u^{n-1} = u(t_{n-1}) = u(t_{n} + \Delta t) = u(t_n) + \Delta t u_t(t_n) + O(\Delta t^2), $$ so you would have $$ \frac{\partial u}{\partial t} + O(\Delta t) $$ on the left side.