Below is Exercise $3.6$ (p.$91$) of "Applied Partial Differential Equations" by Ockendon et al., $2^\mathrm{nd}$ ed.:

Show that, if $$\big(x+\alpha y\big)\dfrac{\partial^2 u}{\partial x^2} + \dfrac{\partial^2 u}{\partial y^2} = 0,\tag{1}\label{1}$$ where $\alpha\neq 0$, the characteristics are given by $$ \left( z\pm \alpha \right)^2 \mp 4\alpha z + 2\alpha^2\log\left(z\pm \alpha \right) + x = \operatorname{constant}, \tag{2}\label{2} $$ where $z^2 = x+\alpha y$. Show that, for small $z$, the characteristics through the origin are $$ \mp \big( x+\alpha y\big)^{\frac32} = \dfrac{3\alpha x}{2}. \tag{3}\label{3} $$ What happens if $\alpha =0$?

First, I do not see how did authors derive expression $\eqref{2}$ for characteristics. For example I tried to carry out proposed change of variables, but the results do not seem to agree with $\eqref{2}$:

Assuming $x$ and $y$ are independent, then the new variable $z = z\left(x,y\right)$ is given implicitly by $z^2 = x+\alpha y$. Differentiating the last expression w.r.t $x$ and $y$ yeilds $$\begin{aligned} \left\lbrace\begin{aligned} z_x = \frac{1}{2z} \\ z_y = \frac{\alpha}{2z} \end{aligned}\right. &\implies \left\lbrace\begin{aligned} \dfrac{\partial u}{\partial{x}} &= \frac{d u}{d z} \!\cdot\!\frac{\partial z}{\partial x} = \frac{1}{2}\frac{ u_z }{z} \\ \dfrac{\partial u}{\partial{y}} &= \frac{d u}{d z}\!\cdot\!\frac{\partial z}{\partial y} = \frac{\alpha}{2} \frac{ u_z }{z} \end{aligned}\right. \\ &\implies \left\lbrace\begin{alignedat}{4} \dfrac{\partial^2 u}{\partial{x^2}} &= \dfrac{\partial }{\partial{x}} \left(\dfrac{\partial u}{\partial{x}} \right) &&= \dfrac{\partial }{\partial{z}} \left(\frac{1}{2} \frac{ u_z }{z}\right) \!\cdot\!\dfrac{\partial z}{\partial x} = \dfrac12\dfrac{zu_{zz} - u_z}{z^2}\!\cdot\! \dfrac{1}{2z} = \dfrac{1}{4z}\dfrac{zu_{zz} - u_z}{z^2} \\ \dfrac{\partial^2 u}{\partial{y^2}} &= \dfrac{\partial }{\partial{y}} \left(\dfrac{\partial u}{\partial{y}} \right) &&= \dfrac{\partial }{\partial{z}} \left(\frac{\alpha}{2} \frac{ u_z } {z}\right) \!\cdot\!\dfrac{\partial z}{\partial y} = \dfrac\alpha2 \dfrac{zu_{zz} - u_z}{z^2}\!\cdot\! \dfrac{\alpha}{2z} = \dfrac{\alpha^2}{4z}\dfrac{zu_{zz} - u_z}{z^2} \end{alignedat}\right.\end{aligned}$$ In this way, assuming $u\left(x,y\right) = u\left(z\right)$ is the solution of $\eqref{1}$, $$\begin{alignedat}{7} z^2=x+\alpha y &\implies {\ \alpha^2 u_{xx} = u_{yy} = \alpha^2z^{-3} \left(zu_{zz} - u_z\right)/4}\\ \eqref{1} & \iff \big(x+\alpha y\big) u_{xx} + u_{yy} = \left(z^2+\alpha^2\right)u_{xx} = \dfrac{z^2+\alpha^2}{4z^3}\big(zu_{zz} - u_z\big) = 0 \\ & \implies u_z=zu_{zz}\implies u_z=Az\implies {u=Az^2+B=A\left(x+\alpha y\right)+B} \end{alignedat} $$

I appreciate any hint regarding where the expression $\eqref{2}$ comes from, as it bothers me most.

Second, can we assume $x\approx y\approx 0$ when talking about characteristics through origin, or does the phrase "for small $z$" imply $ x\approx -\alpha y$ only? Either way I do not see how to get $\eqref{3}$ from $\eqref{2}$.

Third, if $\alpha=0$ then $z^2 = x$. Wouldn't then $\eqref{2} \iff z^2 +x = c$ imply characteristics $x=\operatorname{const}$?

  • $\begingroup$ Maybe I think wrong, but something confuses me: On the one hand we have $z^2=-\frac{u_{yy}}{u_{xx}}|_{u_{xx}\neq 0}$ because of $(1)$ with $z^2=x+\alpha y$, on the other hand your calculations show $\frac{u_{yy}}{u_{xx}}|_{u_{xx}\neq 0}=\alpha^2$ and it follows $x+\alpha y=z^2=-\frac{u_{yy}}{u_{xx}}|_{u_{xx}\neq 0}=-\alpha^2$ and therefore one gets a dependency between $x$ and $y$, which contradicts the assumption, that $x$ and $y$ are independend. I think this must first be resolved before someone can answer your questions. $\endgroup$ – user90369 Apr 11 '17 at 13:45

This answer is not complete (the answer to the first question has an extra $i$), but I am sure it is useful. I hope it can also bring some discussions which I can benefit from. So I decide to post it.

The equation of the characteristics is (refer to Equation 3.20 in Ockendon's book)

$$ \frac{dy}{dx} = \pm \sqrt{-\frac{1}{x + a y}} $$

I only worked on the minus sign. Let $x + a y = z^2$, differentiate w.r.t $x$ gives $1 + a y' = 2 z z'$. The prime means $d/dx$ here. Substitute: $$ 2 z z' = 1 - \frac{a i}{z} $$ Or $$ z' = \frac{z - a i}{2 z^2} $$ This form screams for a solution in terms of $x(z)$, the equation for which is just (the prime means $d/dz$ here, sorry for the abuse) $$ x' = \frac{2 z^2}{z - a i} $$ With help from Mathematica, the result is $$ x(z) = C-a^2 \log \left(a^2+z^2\right)+2 i a^2 \tan ^{-1}\left(\frac{a}{z}\right)+2 i a z+z^2 $$ With $\tan^{-1}(x)=\frac{1}{2i}\ln \left(\frac{x-i}{x+i} \right)$ and some manual simplification we arrive at

$$ x + (a - i z)^2 + 2 a^2 \ln(a+i z) = C $$ This is close to the answer, actually by the replacement $i z \rightarrow z$ we have the answer, unfortunately I don't know how to deal with the $i$.

The second question is much easier, it is just perturbation and Taylor expansion. Setting $x = y = z = 0$ (since we are seeking solution around the origin), we have the "base solution" $$ a^2 + 2 a^2 \ln a = 0 $$ Now put $z$ back and treat it as a small parameter, Taylor expand all terms $$ z^2 + 2 a z + a^2 - 4 a z + 2 z^2 \left( \ln a + \frac{z}{a} - \frac{z^2}{2a^2} + \frac{z^3}{3a^3}\right) + x = 0 $$ It is very interesting that we have to expand up to $O(z^3)$ to have nonzero terms - all lower order terms cancel out. Use the base solution and do the cancellation, we are left with $$ x + \frac{2z^3}{3a} = 0 $$

I am not sure what the 3rd question is asking for. To me the only effect of $a$ is to control the slope of the dividing line between the elliptic and hyperbolic region. With $a = 0$ the dividing line is just the $y$-axis. Please point out if I missed something important.


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.