1

Please help me to solve this problem Let $u(x,y)$ be a solution of a quasi-linear equation of the form:

$a(u_x,u_y)u_{xx} + 2b(u_x,u_y)u_{xy} + c(u_x,u_y)u_{yy}=0$

Introduce new independent variable $\xi, \eta$ and a new unknown function $\phi$ by:

$\xi=u_x(x,y), \eta=u_y(x,y) , \phi=xu_x + yu_y - u $

Prove that $\phi$ as a function of $\xi, \eta$ satisfies $x=\phi_{\xi}, y=\phi_{\eta}$ and linear differential equation

$a(\xi,\eta)\phi_{\eta \eta} - 2b(\xi, \eta)\phi_{\xi \eta} + c(\xi, \eta)\phi_{\xi \xi}$

Here what I did : $\phi=xu_x + yu_y - u $ or $\phi=x\xi + y\eta - u $

Taking derivative $\phi$ w.r.t $\xi$ and $\eta$ we get: $\phi_{\xi}= x$ and $\phi_{\eta}= y$

But then I do not know how to get the linear differential equation $a(\xi,\eta)\phi_{\eta \eta} - 2b(\xi, \eta)\phi_{\xi \eta} + c(\xi, \eta)\phi_{\xi \xi}$

Any help I really appreciate.

Vui Tinh
  • 285

1 Answers1

2

Ended up spending all day on this, but finally understand the solution! (Found this question based on a exercise problem in An Introduction to Nonlinear Partial Differential Equations by J. David Logan)

Anyways, here's the solution: Really, the solution is not difficult, but can be difficult to see or easy to confuse if you're not used to it (as I was).

We know: $u_x=\xi$ and $u_y=\eta$. (ignoring brackets for ease of reading)

From that, we can easily see that $$u_{xx}=\xi_x\\ u_{xy}=\xi_y=\eta_x\\ u_{yy}=\eta_y$$

This is where I had trouble continuing.

This trick is to invert the partials acting on the new coordinates in order to get them instead acting on the old coordinates (Ex. $\xi_x \rightarrow x_\xi=\phi_{\xi\xi}$).

But note that you cannot simply inverse the derivative as you could if there was only one dependent variable($ \xi_x \neq \frac{1}{x_\xi}$), But instead have to complete the matrix inversion (this is really explained well here). (Also, this question expresses changing the partial derivatives in a detailed form)

Therefore, the algebra becomes as follows:

$$ \begin{bmatrix} u_{xx} & u_{xy} \\ u_{yx} & u_{yy} \\ \end{bmatrix} = \begin{bmatrix} \xi_x & \xi_y \\ \eta_x & \eta_y \\ \end{bmatrix} = { \begin{bmatrix} x_\xi & x_\eta \\ y_\xi & y_\eta \\ \end{bmatrix} }^{-1} = { \begin{bmatrix} \phi_{\xi\xi} & \phi_{\xi\eta} \\ \phi_{\eta\xi} & \phi_{\eta\eta} \\ \end{bmatrix} }^{-1} = \frac{1}{\phi_{\eta\eta}\phi_{\xi\xi} -\phi_{\eta\xi}\phi_{\xi\eta}} \begin{bmatrix} \phi_{\eta\eta} & -\phi_{\eta\xi} \\ -\phi_{\xi\eta} & \phi_{\xi\xi} \\ \end{bmatrix} $$

Assuming $u_{xy}=u_{yx}$ and $\phi_{\xi\eta}=\phi_{\eta\xi}$, and letting $\Delta=\frac{1}{\phi_{\eta\eta}\phi_{\xi\xi} -\phi_{\eta\xi}\phi_{\xi\eta}}$. We can see that: $$ u_{xx} = \Delta\phi_{\eta\eta}\\ u_{xy} = -\Delta\phi_{\eta\xi}\\ u_{yy} = \Delta\phi_{\xi\xi}\\ $$

plugging this into the original equation: $$ a(\xi,\eta)\Delta\phi_{\eta\eta} - 2b((\xi,\eta)\Delta\phi_{\eta\xi} + c((\xi,\eta)\Delta\phi_{\xi\xi} = 0 $$ and finally, dividing by $\Delta$, assuming $\Delta\neq 0$

$$ a(\xi,\eta)\phi_{\eta\eta} - 2b(\xi,\eta)\phi_{\eta\xi} + c(\xi,\eta)\phi_{\xi\xi} = 0 $$


For reference, another page I used as reference while trying to figure this out: 1

Dan M.
  • 21