I am a physics student, and have had to use Green's function methods prior (in electrodynamics for example), but things were always badly explained. Now I am trying to brush up on things a bit and figured I'd try to solve some very simple ODEs (that can be otherwise integrated).

Problem is I am not getting sensible results, and I don't even know how can I incorporate boundary conditions into the problem.

Example calculation:

Consider the linear ODE $ f'+\alpha f=g $ where $\alpha$ and $g$ are constants. The linear operator is then $$L=\frac{d}{dx}+\alpha,$$ moreover, since this is a translation-invariant operator, we will have $G(x,x')=G(x-x')$.

Let's assume initial conditions are given by $f(0)=Q$.

Then this equation can be integrated to give $$ f(x)=\left(Q-\frac{g}{\alpha}\right)e^{-\alpha x}+\frac{g}{\alpha}. $$

Now I'll try to solve this ODE using a Green's function method. Let $G(x-x')$ be the Green's function satisfying $$ L_{(x)}G(x-x')=\delta(x-x'), \\ \frac{d}{dx}G(x-x')+\alpha G(x-x')=\delta(x-x'). $$

I have no idea how to proceed, so I'm gonna take a Fourier transform. I have no idea if it is even well defined here or not, but physicists have done worse:

$$ G(x-x')=\int G(k)e^{-ik(x-x')}\ dk, \\ \delta(x-x')= \int\frac{1}{2\pi}e^{-ik(x-x')}\ dk.$$

I get the following for Fourier-components: $$ (\alpha-ik)G(k)=\frac{1}{2\pi}, $$ so $$ G(x-x') =\frac{1}{2\pi}\int\frac{1}{\alpha-ik}e^{-ik(x-x')}\ dk.$$

This integral doesn't converge but I don't worry, as $G$ might just be a singular distribution. Instead, we have $$ f(x)=\int G(x-x')g\ dx'=\frac{g}{2\pi}\int dk\frac{1}{\alpha-ik}e^{-ikx}\int dx'e^{ikx'} \\ =g\int dk \frac{1}{\alpha-ik}e^{-ikx}\delta(k)=\frac{g}{\alpha}. $$

This certainly provides a solution of the equation, but not what I want. And I'm not surprised it didn't work, I didn't specify initial/boundary conditions. Probably when I took the Fourier transform, I accidentally gave some bogus boundary conditions, like $f$ should be zero in infinity or something, but I don't know.

Question: What went wrong? I assume initial/boundary conditions, but how can I control them? How can I specify what initial/boundary conditions do I want? How do I find Green's function for proper initial/boundary conditions? Was taking the Fourier transform the wrong move there?


Green's function method requires homogeneous boundary conditions; that's the only way for solutions to form a linear space. So let's consider the condition $f(0)=0$ for now. To solve the equation $$\frac{d}{dx}G(x-x')+\alpha G(x-x')=\delta(x-x')$$ we need a function $G$ that jumps up by $1$ at $0$ (the derivative of such jump is the Dirac delta), and solves $G'+\alpha G=0$ elsewhere. That is, $G(t)=e^{-\alpha t}H(t)$ where $H$ is the Heaviside function. Subsequently, $$ f(x) = \int_0^\infty g e^{-\alpha(x-x')}H(x-x')\,dx' = g\int_0^{x} e^{-\alpha(x-x')}\,dx' = \frac{g}{\alpha}(1-e^{-\alpha x}) $$

But that was for $f(0)=0$. To solve the original problem $f(0)=Q$, write $f = Q + \tilde f$, where $\tilde f$ solves $\tilde f'+\alpha \tilde f = g-\alpha Q$. So the above applies with $g$ replaced by $g-\alpha Q$, and yields $$ \tilde f(x) = \frac{g-\alpha Q}{\alpha}(1-e^{-\alpha x}) = \left(Q-\frac{g}{\alpha}\right) e^{-\alpha x} + \frac{g}{\alpha}-Q $$ Thus $f=Q+\tilde f$ is exactly the solution you expected.


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.