Let $U_0 \subset \mathbb{R}^3$ be an open neighborhood of $0$ and $X:U_0 \to\mathbb{R}^3$ a smooth vector field, such that $$X(x,y,z) = (X_1(x,y,z),1,0). $$

where $X_1:U_0 \to\mathbb{R}$, satisfies the following hypotheses:

  • $X_1(x,0,z) =0,$ $\forall$ $(x,0,z)$ $\in$ $U_0$.
  • $\frac{\partial}{\partial x}X_1(x,0,z)=\frac{\partial}{\partial y}X_1(x,0,z) =0$, $\forall$ $(x,0,z) \in U_0$
  • $\frac{\partial}{\partial y}X_1(x,0,z)\neq 0$ $\forall (x,0,z) \in U_0$
  • $X_1(0,0,0)=(0,0,0)$.

I need to find a change of coordinates $\varphi: W_0\subset \mathbb{R}^3 \to V_0\subset \mathbb{R}^3$ ($\varphi(0)=0$), such that $$Z(x,y,z) = \text{d}\varphi_{\varphi^{-1}(x,y,z)}X(\varphi^{-1}(x,y,z)) = (y,1,0) $$ or $$Z(x,y,z) = \text{d}\varphi_{\varphi^{-1}(x,y,z)}X(\varphi^{-1}(x,y,z)) = (-y,1,0). $$

The problem that I am facing does not allow me to mess up the z and y-axes in my coordinate system.

A natural way to solve this problem is trying to find this changing of coordinates in the following form $$\varphi(x,y,z) = (f(x,y,z),y,z).$$

(with $\frac {\partial f}{\partial x} (0) \neq 0$ and $f(0,0,0)=0$). If this coordinate change works, we would get

$$Z(\varphi(x,y,z)) = \text{d}\varphi(x,y,z) \cdot X(x,y,z)$$ $$(\pm y,1,0) = \left( \frac{\partial f}{\partial x}(x,y,z) X_1(x,y,z) + \frac{\partial f}{\partial y} (x,y,z) ,1,0 \right) .$$

And then my question appears, does anyone know if this PDE has a solution?

$$ \frac{\partial f}{\partial x}(x,y,z) X_1(x,y,z) + \frac{\partial f}{\partial y} (x,y,z)= \pm y,$$ $$f(0,0,0)=0,$$ $$\frac{\partial f}{\partial x}(0,0,0) \neq 0. $$


The main idea is to use the characteristic method. It is motivated by the following equation: $$\langle \left(\nabla f\circ \gamma\right)(t), \gamma'(t) \rangle = g(t).$$ The curve $\gamma$ is called the characteristic cuve.

The first equality follows from the chain rule: remember that:

$$\frac{d}{dt}f\circ \gamma(t) = \langle \nabla f\circ \gamma(t),\gamma'(t)\rangle. $$

Note that $$\langle \left(\nabla f\circ \gamma\right)(t), \gamma'(t) \rangle = g(t) = \sum_{j=1}^3\frac{\partial f}{\partial x_i}\frac{dx^i}{dt},$$ therefore, one convert the PDE on a system of ODE's. After all, we isolate $t$ to recover the dependence of the coordinates.

Therefore, one must search for solutions for the problem:

$$\frac{dx}{dt} = X_1(x,y,z),$$ $$\frac{dy}{dt} = 1,$$ $$\frac{df}{dt} = \pm y,$$ and $\frac{dz}{dt}$ can be free.

Then, one has:

$$y(t) = t + c,$$ so $$\frac{df}{dt} = \pm (t+c),$$ and therefore, $$f(t) = \pm(\frac{t^2}{2} + ct) + d.$$

Once $t = y - c,$ we have:

$$f(x,y,z) = \pm((y-c)^2 + c(y-c)) + d.$$

Now, we want to relate our equation with $X_1.$ Once $t = y -c$, one has that $\frac{dx}{dt} = \frac{dx}{dy}$. Therefore, we have:

$$\frac{dx}{dy} = X_1(x,y,z).$$ This is a non autonomous ODE. To finish, convert this equation on an autonomous equation and impose your initial data, this must determine $x$ in terms of $y$. Use inverse theorem to obtain your function $f$ in terms of $X_1$. I can add more details if needed.

Note that the solution we have obtained if a function of $y$. Since you want to obtain a solution in terms of $x$ you could follow two ways:

$1)$ Instead of isolating $t = y-c$ you could just to plug $y = t+c$ on the equation for $x$, obtaining:

$$\frac{dx}{dt} = X_1(x,t+c,z).$$

This is an non autonomous ODE. You will obtain a solution $x(t)$. Therefore, you must isolate $t$ in terms of $x$, and then, plug this on the expression for $f(t)$ to obtain $f$ in terms of $x, X_1.$

$2)$ Follow the way I suggested at first. Solve the equation for $x(y)$. This is what I meant at first. But $1)$ seems more natural, although it leads to the same answer.


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.