non uniqueness for the solution of the following Cauchy problem

148 Views Asked by At

The following Cauchy Problem:

$$y'=-10 \sqrt y,~~~ y(0) =\dfrac 14$$

doesn't have a unique solution, is that correct? I tried to solve that, and the solution is a parabola, but what I think is that when the parabola reach zero then there is also another solution which is the first part of the parabola and then the solution $y=0$.

The solution for the diff. eq. is $y(x) = \frac14(1-10x)^2$, but when $x=1/10$ the parabola reaches zero and then there is another solution which is $y(x) = \frac14(1-10x)^2$ for $x<1/10$ and $y=0$ for $x\ge1/10$.

Moreover I need to compute the absolute stability interval for Forward Euler applied to the previous ODE. I know how to compute it for the test equation (linear case), but I don't know how to deal with this particular case.

1

There are 1 best solutions below

10
On

Looking at the equation one can note first that

  • $y$ has to be always non-negative and
  • the slope $y'$ is always negative or zero for $y=0$.

This means that the solution that is initially parabolic gets then absorbed by the $x$-axis, there can be no solution that follows the rising parabolic branch after touching the $x$-axis.


A-stability of the Euler method for an ODE $y'=f(y)$ means roughly that if $\lambda=f'(y_n)<0$, and $y_n$ is close to a fixed point or an asymptote of the equation, then one wants that $λh_n=h_nf'(y_n)\in (-2,0)$, better closer to zero than the other side.

Here $f'(y)=-5y^{-\frac12}$ so that $0<h_n<\frac25\sqrt{y_n}$ as a guideline. With a fixed step size this bound is eventually violated. Let's see what happens in variable step size.

Selecting for instance $h_n=\frac15\sqrt{y_n}$, this gives the iteration $$ y_{n+1}=y_n+h_nf(y_n)=y_n-2y_n=-y_n $$ which is obviously not what is desired. So the guideline seems only useful if the local Lipschitz constant can be bounded.

Let's try a smaller $h_n=\frac1{20}\sqrt{y_n}$, then $$ y_{n+1}=y_n-\frac12y_n=\frac12y_n\implies y_n=2^{-n}y_0. $$ It follows that $h_n=\frac1{20}2^{-n/2}\sqrt{y_0}$, but $$ t_n=h_0+h_1+\dots+h_{n-1} =\frac1{20}\frac{1-2^{-n/2}}{1-2^{-1/2}}\sqrt{y_0} <\frac{1+2^{-1/2}}{10}\sqrt{y_0}, $$ so that the time steps can not move the time over this finite boundary, which is also not what is expected of a solution.