Method of Characteristics for a non-linear PDE

270 Views Asked by At

I've been trying to work through some of the more difficult questions we've been given in class in regards to the method of characteristics for solving PDEs, but I've come a bit unstuck.

I've been asked to solve the following question;

$$x^2 u_x + e^{-y} u_y = 2y u^2$$ Subject to Cauchy data $u(s) = e^{-s}$ along $(x(s), y(s)) = (s,1)$

So, firstly, I set up the following ODEs; $$\frac{dx}{dt} = x^2, \frac{dy}{dt} = e^{-y}, \frac{du}{dt} = 2y u^2$$

Solving these three ODEs, I acquired the following expressions for $x$, $y$ and $u$... $$ x = \frac{1}{A - t}, y = \ln(t + B), u = \frac{1}{C - 2yt}$$ Where $A, B, C$ are all constants.

From there, I used my Cauchy data, and set $x = s$, $y =1$, and $u = e^{-s}$. Then, I set $t = 0$, in order to determine what my constants were.

I then got $A = \frac{1}{s}$, B = $e$, and $C = e^s$.

Now, rearranging my equation for $y$, I found that $t = e^y - e$, and using this, coupled with a rearrangement of my equation for $x$, I found an expression for $s$; $$s = \frac{x}{x(e^y-e)+1}$$.

Then, all that remained was to put everything into my expression for $u$; $$ u(x,y) = \frac{1}{e^{\frac{x}{x(e^y -e)+1}} - 2 y (e^y - e)}$$

However, this seems way too complicated,and I feel like I've made a mistake somewhere in my working.

I feel like my $\frac{du}{dt} = 2yu^2$ logic is kind of incorrect, and that I should be replacing the $y$ term here with $\ln(t + B)$, but even then, the integration becomes really horrible.

If anyone could provide some input, that would be fantastic. :)

EDIT: Yeah, I had made a mistake.

So, my equations and working for my $x$ and $y$ terms were fine, but my $u$ was bad.

To refresh, I've got that $y = \ln(t + e)$. Then,

$$\frac{du}{dt} = 2 y u^2 = 2 \ln(t+e) u^2$$ $$ \frac{du}{u^2} = 2 \ln(t+e) dt$$ $$ \frac{1}{u} = -2(t+e)\ln(t+e) + 2t + C$$

Now, using my Cauchy data, I have that $u = e^{-s}$. Setting $t=0$, we have... $$\frac{1}{e^{-s}} = -2 (0 + e) \ln(0+e) + 2 \times 0 + C$$ $$ C = e^s - 2e$$

Then, using some manipulation of the equation I got for $y$, I can write $1/u$ as follows; $$\frac{1}{u} = e^s - 4e + 2e^y (1 - y)$$

From here, I am assuming that I need to rearrange my equation for $x$ to make $s$ the subject, and then place that into the $e^s$ term, but still, this seems almost over the top. Is that really all I need to do??

1

There are 1 best solutions below

2
On BEST ANSWER

Though it looks complicated, the solution is

$u(x,y) = \dfrac{1}{e^{\frac{x}{(e^y-e)x+1}} - 2 e^y(y -1)}$, which can be verified by substituting into the PDE.

The initial condition of the PDE is given by $u(s) = e^{-s}$ along $(x(s),y(s))=(s,1)$, which is $u_0(x,y) = u(x,1)=e^{-x}$.

The characteristic ODE are

$\dfrac{d x}{d t} = x^2$, $\dfrac{d y}{d t} = e^{-y}$, $\dfrac{d u}{d t} = 2 y u^2$,

with the initial conditions $x(0)= x_0$, $y(0)=1$ and $u(0) = u_0$ at $t=0$.

By solving the first two ODE, we obtain $x = \dfrac{x_0}{1-x_0 t} $ and $y = \ln (t + e )$.

By substituting $y$ into the third ODE, we obtain $\dfrac{d u}{d t} = 2 \ln (t + e ) u^2$ with initial $u(0)= u_0$.

Solve this ODE, then, $u = \dfrac{u_0}{1- 2 u_0 (t+e) (\ln (t+e) - 1) }$.

Obtain $x_0$ and $t$ from $x$ and $y$ respectively, and substitute $x_0$, $t$ and $u_0$ into $u$.

Thus, the solution is obtained, which is $u(x,y) = \dfrac{1}{e^{\frac{x}{(e^y-e)x+1}}-2 e^y(y -1)}$.