3
$\begingroup$

I just knew how to do Newton-Raphson iteration in time-independent 1D nonlinear differential equation. Then I applied to time-dependent 1D nonlinear differential equation, and I got confused.

Below is just test time-dependent 1D nonlinear differential equation I want to solve (I just made this up from heat equation, to make it nonlinear) $$ \frac{\partial u}{\partial t}+\frac{\partial^2 u}{\partial x^2}+u^2=0 $$ For this equation, I assumed I would need initial condition, and boundary condition $$ u(x,t=0)=\left\{ \begin{aligned} 1\qquad&(-1\leq x\leq 1)\\ 0\qquad&\text{otherwise} \end{aligned}\right.,\qquad u(-5,t)=0,\quad u(5,t)=0 $$ I set the $x$ range of $-5 \leq x \leq 5$, and planned to create code in Matlab.

I started with space discretization by using second-order difference $$ \frac{\partial u}{\partial t}+\frac{u_{i+1}-2u_{i}+u_{i-1}}{\Delta x^2}+u_{i}^2=0 $$ where $i$ denotes space discretization number

Then I applied Backward-Euler discretization for time discretization, which is $$ \frac{d q(x,t)}{dt}=f(t,x,q(x,t))\qquad\longrightarrow\qquad \frac{q_{j+1}-q_{j}}{\Delta t}=f_{j+1} $$ where $j$ denotes time discretization number (this is just intermediate step for me. I am planning to apply Trapezoidal Rule - Second order Backward Difference Formula (TR-BDF2) later) Now applying this, the equation looks like $$ \frac{u_{i,j+1}-u_{i,j}}{\Delta t}+\frac{u_{i+1,j+1}-2u_{i,j+1}+u_{i-1,j+1}}{\Delta x^2}+u_{i,j+1}^2=0 $$ Again, $i$ and $ j$ denotes space and time discretization respectively.

Then I applied Newton-Raphson(NR) scheme, which is for give equation $$ F(u)=0 $$ the solution is determined by iteration $$ F'(u^{k})\delta u^{k}=-F(u^{k}), \qquad \qquad u^{k+1}=u^{k}+\delta u^{k} $$ (I'm running out of space for discretization) where k denotes NR iteration number, and if $F(u)$ was system of equations, $F'(u)\delta u$ should be Jacobian. Do I have to set initial and boundary condition for $\delta u_{i,j}$ as well?

Now my equation looks like $$ \frac{\delta u_{i,j+1}^{k}-\delta u_{i,j}^{k}}{\Delta t} + \frac{\delta u_{i+1,j+1}^{k}-2\delta u_{i,j+1}^{k}+\delta u_{i-1,j+1}^{k}}{\Delta x^2} +2u_{i,j+1}^{k}\delta u_{i,j+1}^{k} = -\frac{u_{i,j+1}^{k}-u_{i,j}^{k}}{\Delta t} -\frac{u_{i+1,j+1}^{k}-2u_{i,j+1}^{k}+u_{i-1,j+1}^{k}}{\Delta x^2} -(u_{i,j+1}^{k})^2 $$

This is where I got stuck.

How do I proceed after this?

Am I missing something? like another boundary/initial conditions?

Because I only know $u_{i,1}^{0}$ and $\delta u_{i,1}^{0}$.

There are like, 11 unknowns due to backward differences ($j+1$).

Some literature says that I should solve this equation for "each time step"

Or, is this right way to apply Backward-Euler scheme and NR scheme to the time-dependent nonlinear differential equation?

Is there any good example solving time dependent nonlinear differential equation with Newton-Raphson iteration?

$\endgroup$
  • $\begingroup$ Remember that the free variable in time dependent problems is the result in the future, not the current tilmestep. Your last expansion only needs to be in (in your notation $u_{i,j+1}$). Also your boundary condition applies at all times, not just the past. $\endgroup$ – origimbo Apr 1 '16 at 14:08
2
$\begingroup$

You seem to be confused with what are unknowns in your algebraic equations (as mentioned also in the comment by origimbo).

Let say you have $N$ inner nodes for space discretization, i.e. $\Delta x = 10/(N+1)$. The values $u_{i,1}$ shall be obtained from initial condition (a more usual notation is $u_{i,0}$).

The notion "you have to solve your algebraic equations in each time step" means that in each time step the values $u_{i,j}$ are known (either from initial condition or from previous time step) and your unknowns (the values to be found) are only $u_{i,j+1}$. Now this is the task for NR method.

When applying NR method, you have to choose the values $u^0_{i,j+1}$, typically you take $u^0_{i,j+1}=u_{i,j}$. It means $F(u)=0$ represents $N$ nonlinear algebraic equations having $N$ unknowns $u_{1,j+1}$ up to $u_{N,j+1}$ having the form $$ \frac{\delta u_{i,j+1}^{k}}{\Delta t} + \frac{\delta u_{i+1,j+1}^{k}-2\delta u_{i,j+1}^{k}+\delta u_{i-1,j+1}^{k}}{\Delta x^2} +2u_{i,j+1}^{k}\delta u_{i,j+1}^{k} = -\frac{u_{i,j+1}^{k}-u_{i,j}}{\Delta t} -\frac{u_{i+1,j+1}^{k}-2u_{i,j+1}^{k}+u_{i-1,j+1}^{k}}{\Delta x^2} -(u_{i,j+1}^{k})^2 $$

In above, the values $u_{i,j}$, $u_{*,j+1}^k$ are known, the unkowns in above are only three (but in each equation different ones), namely $\delta u_{i-1,j+1}^k$, $\delta u_{i,j+1}^k$, and $\delta u_{i+1,j+1}^k$. The boundary conditions are $\delta u_{0,j+1}^k=0$ and $\delta u_{N+1,j+1}^k=0$.

By the way, the sign in your PDE before the diffusion term is wrong, it should be minus.

$\endgroup$

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.