6
$\begingroup$

Is it possible to solve a differential equation of the following form?

$\partial_x^2y + \delta(x) \partial_x y = 0$

where $\delta(x)$ is the dirac delta function. I need the solution for periodic boundary conditions from -1 to 1, but I'll be fine if you direct me for any sort of boundary conditions..

I've realised that I can solve this for some types of boundary conditions. What i'd be really interested in is how to do this for periodic boundary conditions...

Technically, if I approach the problem by splitting the regions $x<0$ and $x>0$ and solve in each part separately, I can solve it and get linear equations in both regions. This will give me $4$ variables. Periodicity, and periodicity of the derivative will give me 2 equations. Continuity at $x=0$ will give me one more. How do i relate the derivative around the $x=0$ interface?

A little background: If there was no delta function, but rather say some gaussian approximation, I would be expect to be able to solve it, but I don't see why I can't get the information of the derivative around $x=0$ when i put in a dirac delta function. My actual problem is reasonably more complicated but this is the quickest simple example I could reduce my problem to. If I try to integrate in an epsilon region around $0$, then I end up with an expression in $y^\prime(0)$, which isn't defined.

If it helps, i'm actually interested in a physical problem, so this delta function is just an approximation, but i'm unable to take any function in it's place whose limit I can make tend to a delta function..

Any help or direction would be greatly appreciated!

EDIT: I guess I should make my actual problem a bit clearer as well. In this simplistic case i can simplify the differential equation by means of the substitution of $z=y^\prime$, however I can't really do that in my original equation. I'm basically interested in some technique by which I can get the information for the change in derivative of the function around the delta function. A better example might be $\partial_x^2y + \delta(x) \partial_x y +y= 0$

$\endgroup$
  • $\begingroup$ The equation you've written reduces to $y''(x)=-y'(0)\delta(x)$, since $y(x)\delta(x)$ is supported, as a distribution, at the origin. $\endgroup$ – Jonas Dahlbæk Jul 8 '14 at 13:12
  • $\begingroup$ The new equation reduces to $y''+y=-y'(0)\delta$, which is probably a better candidate for periodic boundary conditions. $\endgroup$ – Jonas Dahlbæk Jul 8 '14 at 13:18
  • 1
    $\begingroup$ True, but strictly speaking, I don't expect $y^\prime(0)$ to be defined, and I can't treat it like an arbitrary constant. I only expect $y(x)$ to be continuous at $0$. Normally in such scenarios you integrate in an $\epsilon$ neighborhood of the interface to obtain relations between the derivatives at both sides, but that isn't going to work here.. So how do I proceed? $\endgroup$ – SarthakC Jul 8 '14 at 13:18
3
$\begingroup$

Set $z=\partial_x y$. Then $z=z(x)$ satisfies the equation $$ \partial_x z+\delta(x)\,z=0, $$ with solution $$ z(x)=z(0)\,\mathrm{e}^{-\int_0^x\delta(\xi)\,d\xi}. $$ Hence, in order to obtain $y=y(x)$ you should integrate $z$ once.

$\endgroup$
  • $\begingroup$ Um.. I guess you're right. But could you tell me how to approach this for the case of periodic boundary conditions? I've edited the question accordingly. I realize this can be solved for some cases, but can it be solved for periodic boundary conditions? $\endgroup$ – SarthakC Jul 8 '14 at 13:05
1
$\begingroup$

We can reduce to a first order equation: $$ e^{ix}(i\partial_x)e^{-2ix}(-i\partial_x)e^{ix} y=y'(0)\delta. $$ Now, since the RHS is supported at the origin, we obtain $$ \partial_x [e^{-2ix}(-i\partial_x)e^{ix} y]=-iy'(0)\delta, $$ or $$ \partial_x [e^{-2ix}(-i\partial_x)e^{ix} y+iy'(0) H]=0, $$ where $H$ is the heaviside step function. It follows that $$ e^{-2ix}(-i\partial_x)e^{ix} y+iy'(0) H=c $$ for some constant $c\in\mathbb C$. Thus, $$ \partial_x e^{ix}y(x)=ie^{2ix}(c-iy'(0)H(x)). $$

The problem, then, is that this equation implies that $y'$ has a step-function type singularity at the origin, where it jumps an amount $y'(0)$. But $y'(0)$ is not well-defined then, and thus neither is our starting equation! This stems from the term $y'\delta$. If you expect $y'$ to not be continuous at $0$, it doesn't make sense to multiply the delta-distribution with $y'$, since $$ (y'\delta)(g)=\delta(g y')=y'(0)g(0). $$ Instead, let us formulate a different equation, $$ y''+y=-C_y\delta $$ for some $C_y\in\mathbb C$ independent from $y'(0)$, say for instance $C_y=(y'(0+)+y'(0-))/2$. For a continuous function $y$, $C_y=y'(0)$, but in your case $C_y$ is well-defined even though $y$ is not assumed continuous. Other choices would be possible candidates as well.

$\endgroup$
  • $\begingroup$ Yes, but the derivative would NOT be continuous at $x=0$, right? If it were, then $y^{\prime\prime}$ would have at most a jump discontinuity, and hence the LHS would have a jump discontinuity while the RHS is a delta function.. $\endgroup$ – SarthakC Jul 8 '14 at 13:23
  • $\begingroup$ The idea is that I expect there to be a change in the derivative around $x=0$. What i don't get is how to quantify it. As you've done, the change in derivative turns out to depend on $y^\prime(0)$, which I don't know/isn't defined.. $\endgroup$ – SarthakC Jul 8 '14 at 13:25
  • $\begingroup$ Hmm. Could you clarify on your last step? Why is the RHS zero? Ideally wherever you have written $f_\epsilon$ there should be a limit of $\epsilon \rightarrow 0$, right? And in the last step you swap the limits of $x\rightarrow 0$ and $\epsilon \rightarrow 0$. I don't think that is justified, because your final result is saying that $ey$ is differentiable at $0$, which would be equivalent to saying that $y$ is differentiable at $0$ (since $e$ is just $e^{ix}$) function (am I right in saying that?). But we know $y$ is not. I think swapping the limits over there might be unjustified.. $\endgroup$ – SarthakC Jul 8 '14 at 14:10
  • $\begingroup$ Well. $y^{\prime\prime}$ was integrable, right? Just discontinuous. I believe the problem earlier was the swapping of the limits. In your new edit, I don't quite see how to proceed from where you've left to getting the final solution.. Could you just point a few more steps? What do I do with the $y^\prime(0)$ term? If i integrate your last equation from $-1$ to some arbitrary $s$, and apply continuity at $s=0$ on $y$, then that will give me $y^\prime(0)=0$, and $y$ continuous and differentiable everywhere. Is that even correct? $\endgroup$ – SarthakC Jul 8 '14 at 14:29
  • $\begingroup$ Umm.. Yes, I already got that $y(x)$ has a ramp function type singularity(by which you mean something like $|x|$, right?). But thanks for showing this more rigorously. However, I'm still no where closer in getting the solution, right? Being able to get rid of the $y^\prime(0)$ term has already been my headache for a long time... $\endgroup$ – SarthakC Jul 8 '14 at 15:01

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.