Take a Klein-Gordon (KG) equation for a model exercise: \begin{equation}\frac{\partial^2 u}{\partial t^2}=c^2\frac{\partial^2 u}{\partial x^2 } - \Omega^2 u,\end{equation} with boundary and initial conditions: \begin{equation}u(x,t)\in[0,a]\times[0,\infty]\end{equation} \begin{equation}u(x,0)=\alpha(x)\end{equation} \begin{equation}\frac{\partial u}{\partial t}(x,0)=\beta(x)\end{equation} \begin{equation}u(0,t)=0\end{equation} \begin{equation}u(a,t)=0\end{equation} So nothing special until here.

I've learnt two methods for this, one of the is separation of variables, which nicely gives the dispersion of the wave as $\omega^2 = c^2 k^2 + \Omega^2$ and its solution as a Fourier series: \begin{equation}u(x,t)=\sum_n\sin (k_n x)\left[A_n\sin(\omega_n t) + B_n\cos(\omega_nt)\right]\end{equation}

The second method starts by supposing the solution can be written in the form $e^{-i\omega t}e^{ikx}$ the minus sign being there because of convention.

Substitituing this into the KG equation I get $(-i\omega)^2 = c^2(ik)^2 - \Omega^2$ which gives the dispersion of the wave, but then I just don't know how to proceed.

If I express $\omega_{\pm}=\pm\sqrt{c^2 k^2 + \Omega^2}$, and wite the solution as $u(x,t)=\int dk e^{ikx}\left(A(k)e^{i\omega_+t}+B(k)e^{i\omega_-t}\right)$, I reach a dead end, and can not go on to write the solution as a Fourier series. Furthermore I get (from the initial conditions): \begin{equation}u(x,0)=\int dk e^{ikx}(A(k)+B(k)) = \alpha(x) \rightarrow A(k)+ B(k) = \frac{1}{2\pi}\int dx \alpha(x)e^{-ikx}\end{equation} \begin{equation}\dot{u}(x,0)=\int dk e^{ikx}\omega(A(k)+B(k)) = \beta(x) \rightarrow \omega[A(k)+ B(k)] = \frac{1}{2\pi}\int dx \beta(x)e^{-ikx}\end{equation} Thus $\omega = \frac{\int dx \beta(x)e^{-ikx}}{\int dx \alpha(x)e^{-ikx}}$, which is rather strange. Anyway, I don't know how to proceed from here.

If I go the other way around, and express $k_{\pm}=\pm\sqrt{\frac{\omega^2-\Omega^2}{c^2}}$, then this will be $e^{i\omega t}(Ae^{ik_+x}+Be^{ik_-x})$ (I don't know why I don't integrate here, but if I do, nothing will come out of this). Using the boundary conditions, I get $u=\sum_n2iA_nsin(k_nx)e^{i\omega t}$ with $k_n=n\pi/a$ as expected, but again I don't know what to do from here, because this gives a completely different Fourier series than the form of the solution (not to mention it is complex).

Please point out what am I doing wrong.


I believe you are forgetting the dispersion relation i.e. that $\omega$ is a function of $k$ when you work out $\dot{u}(x,0)$ as well as a sign. I make the calculation to be as follows. Since:

$$\begin{array}{lcl}u(x,t)&=&\int_0^\infty \,\mathrm{d} k\, e^{i\,k\,x}\left(A(k)e^{i\omega_+t}+B(k)e^{i\omega_-t}\right)\\&=&\int_0^\infty \,\mathrm{d} k\, e^{i\,k\,x}\left(A(k)\,\exp(i\,\sqrt{c^2 k^2 + \Omega^2}\,t)+B(k)\,\exp(-i\,\sqrt{c^2 k^2 + \Omega^2}\,t)\right)\end{array}\tag{1}$$


$$\dot{u}(x,t)=\int_0^\infty \,\mathrm{d} k\, e^{i\,k\,x}\,\sqrt{c^2 k^2 + \Omega^2}\,\left(A(k)-B(k)\right)\tag{2}$$

so you don't get your "strange" $\omega$ formula: there is no one, unique $\omega$ and instead you have a non sinusoidal superposition.

You're also forgetting the two $k$-branches branches in your last paragraph as well as the fact that $\omega$ must be allowed to vary over positive and negative values: the general solution for a given $\omega$ is $e^{i\,\omega\,t}\,(A_n\,e^{i\,k(\omega)\,x} + B_n\,e^{-i\,k(\omega)\,x})$, where $k(\omega)= \sqrt{\frac{\omega^2-\Omega^2}{c^2}}$. To fulfil the boundary conditions, you must have $B_n=-A_n$ and of course $A_n$ can be generally complex, so write $B_n=-A_n = i\,C_n$ and you will get your $u=\sum_n 2\,i\,A_n\,\sin(k_n\,x)\,e^{i\,\omega\,t}$. Now you recall that there are two $\omega$-branches, so for every $k_n=n\pi/a,\,n\in\mathbb{N}$ there are two $\omega$s: $\omega = \pm\sqrt{c^2 \left(\frac{n\pi}{a}\right)^2 + \Omega^2}$. If you like, you can think of the negative $\omega$ solutions as being associated with $-n$ values: hence you get your most general possible superposition:


for arbitrary $D_n\in\mathbb{C}$, whence you can derive the first Fourier series you cited, and indeed you can find complex $D_{\pm n}$ to make the series real-valued.

| cite | improve this 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.