0
$\begingroup$

How to solve second order non-linear ODE $$y'^2+y y''+\frac{2}{x} y y' -0.1 y^2=0$$ subject to $y'(1)=0$ and $y(1)=1$ over the interval $0 < x \le 1$.

I turned the equation to a PDE $y'^2+y y''+\frac{2}{x} y y' -0.1 y^2=y'_t$. I was trying to find the steady state solution when $t \to \infty$, which is the solution to the ODE. I used an explicit finite difference scheme in MATLAB. But it doesn't seem to give the right solution.

I have problem implementing boundary conditions in the MATLAB.

I will be grateful if you help me solve this. Also mentioning any other numerical method will be great. Thanks.

|cite|improve this question|||||
$\endgroup$
  • $\begingroup$ Cross-posting is discouraged on the StackExchange network, so people don't waste their time with an answer you already received on the other site. The usual procedure is to wait a few days, and then either raise a flag and ask the moderators for migration (if there are some answers already) or delete the old and ask a new question. math.stackexchange $\endgroup$ – Mauro Vanzetto Jan 29 '18 at 16:28
  • $\begingroup$ This is an initial value problem, so why are you not just using Matlab's built-in ODE solvers like ode45? $\endgroup$ – Kirill Jan 29 '18 at 18:37
2
$\begingroup$

You have two conditions at x=1, I hope this is correct in which case you have an initial value problem. If yes, then you can write this as a first order system by introducing $z=y'$

$$ z' = y/10 - z^2/y - 2z/x, \qquad y' = z $$ with initial condition $$ y(1) = 1, \qquad z(1) = 0 $$ Solve this using an ode solver from $x=1$ to $x=0$. As you approach $x=0$, you will face the problem of division by zero, unless $z \to 0$. There is also possibility of $y$ becoming zero. Do you have a reason to expect your equation should have non-negative solutions ?

|cite|improve this answer|||||
$\endgroup$
  • $\begingroup$ I have doubt about boundary conditions. I wanted to see if the conditions are reasonable or not. I solved this analytically using MATHEMATICA. It gives: $$y=(1+i)\exp\left(\frac{1}{10}\left(-5\log(x)+5\log\left(\cosh\left(x/\sqrt{5}+0.2\left(\sqrt{5}+5\text{arctanh}\left(\sqrt{5}\right)\right)\right)\right)\right)\right)$$ which has a very small imaginary part and goes to $+ \infty$ at $x \to 0$. From this complex number I doubt that my friend has given me the wrong boundary conditions! $\endgroup$ – Ghartal Jan 29 '18 at 12:19
1
$\begingroup$

Sometimes, pdes result in solutions that are not steady (flow around a cylinder) and therefore you face a problem in which infinitely many solutions arise. This kind of problems are not suitable for temporal discretisations if one particular solution must be reached. I mean the limit $$\lim_{t\to \infty}y(t)$$ does not make any sense.

Instead you can linearise the problem (Newton's method) by setting: $$y^{n+1}=y^n+\delta y^n$$ And solve for $\delta y^n$ from linear problems carrying a substitution into your nonlinear model. Apply homogeneous dirichlet BC.

Keep in mind that nonlinear terms in $\delta y^n$ can be neglected because it is supposed small.

If the funcion $y$ is smooth enough near the possible solutions Newton's method will do the job with quadratic convergence.

|cite|improve this answer|||||
$\endgroup$
  • $\begingroup$ Thanks. Newton will do it. I have problems writing the Matlab code. Thank you again. $\endgroup$ – Ghartal Jan 29 '18 at 12:12

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.