For a non-vanishing field $X$ and $a\in C^1(\mathbb R^n;\mathbb C)$, how to reduce the study of $X\cdot \nabla u +au= f$ to the study of $Xv=g$?

117 Views Asked by At

I'm doing exercises in the first chapter of Serge Alinhac's "Hyperbolic Partial Differential Equations", and while most of the other exercises for this chapter are straightforward, I'm stuck on Q3:

Let $X$ be a non-vanishing field in $\mathbb R^n$ and $a\in C^1(\mathbb R^n)$ a complex function. Explain how the study of the equation $Xu+au = f$ can be (locally) reduced to the study of the equation $Xv=g$.

Here, Serge is using the notation $Xu \triangleq X\cdot\nabla u$; $u$ is a scalar function (presumably, due to $a$, taking values in $\mathbb C$). I would assume $f,g$ can be taken to be smooth.

Things I've tried:

  • The "(locally)" that appears in the question suggests to me a change of variables. But if I say let tildes mean composition with some nice $\phi$, e.g. $\tilde u=u\circ \phi$ and so on, all I end up with is $(\tilde X \cdot ((\nabla \phi)^{-1}\circ \phi)\nabla \tilde u) + \tilde a \tilde u = \tilde f$ which doesn't seem helpful.

  • I tried to consider the real and imaginary parts. (Earlier in the book, it is remarked that $\partial_t + i\partial_x$ is the Cauchy-Riemann operator; the real and imaginary parts of $\partial_tu + i\partial_xu =0$ form precisely the standard system called the Cauchy-Riemann equations) But I don't see any way to continue from the resulting system. If say $u=v+iw$, $f=g+ih$, and $a=b+ic$, then the two equations you get for $(v,w)$ are $$ Xv+bv - cw = g, \\ Xw + cv + bw = h.$$

  • I tried to use what I know from basic ODE theory, namely integrating factors. If I could solve $Xz = az$, with $z\neq 0$ everywhere, then by Leibniz rule, $$ X (uz) = zXu + uX z = z Xu + au z $$ Then $$Xu + au = f\iff zXu + auz = fz \iff X(uz) = fz$$ Of course, one takes $uz=v$ and $fz=g$, to match the question's notation, so this feels like the most promising approach. But I don't get what the complex part of $a$ is supposed to do for me, and the book did not consider solving things like $Xz = az$ yet, nor am I 100% sure its always possible to solve this equation.

Are any of these a step in the right direction, and does anyone have some pointers?


(edit after sleeping) So in the case where $X=\partial_1$ and $a$ takes purely imaginary values, it works out like this. You start the solution of $Xz=az$ at whatever $z_0$ say $z_0=1$. The solution is $z(x) = z_0 \exp(\int_0^{x_1} a(s,x_2,\dots,x_n) ds)$. My stronger assumption on $a$ means that $$\partial_1\left( \frac{|z|^2}2 \right) = \Re(\partial_1 z\overline z)= \Re ( a |z|^2) = 0$$ so that $|z|^2$ is a constant, and in particular $z\neq 0 $ everywhere.

My lack of formal training in differential geometry is showing, but I suppose from try #1 above, the nonvanishing of $X$ allows me to reduce to this case? I guess I need to solve $\partial_1 = \tilde X \cdot ((\nabla \phi)^{-1}\circ \phi)$?

And also, my hunch now is that the result is false if $a$ can take more general complex values...

1

There are 1 best solutions below

3
On BEST ANSWER

I suppose that Alignac provides details for showing that $Xv=g$ admits smooth local solutions in a neighborhood of a point where $X$ is non-zero? (e.g. by integrating along the flow of X).

After that you pose and solve locally, as suggested in comments, $XA=a$ and then $XB=e^{A}f$ (there is no need to separated into real and imaginary parts). After that $u=Be^{-A}$ provides a local solution to the problem since $$ X u + a u = (X + a) (B e^{-A}) = (XB) e^{-A}-B a e^{-A} +a B e^{-A} = f $$

The non-vanishing of $X$ ensures that the problems are equivalent. Otherwise you run into counter examples like: $$ x u'(x) + u(x) = 1 $$ which does have a smooth solution ($u\equiv 1$) but for which $x A'(x)=1$ does not have a solution in a neighborhood of zero.