I am playing with the partial differential equation $\frac{\partial^2 \phi}{\partial x^2} - \frac{\partial^2 \phi}{\partial t^2} = \lambda |\phi|^2 \phi$. $\phi(x,t)$ is complex and the domain is not bounded.

I would like to see something like a free-wave solution and perhaps investigate its dispersion relation. The left-hand side is just a common wave equation, being satisfied with any function of the form $f(x \pm t)$. The right hand side will produce the interesting effects.

I tried the following. Let's assume $\phi(x,t) = A(x,t) e^{i(\omega t - kx)}$. Then plug it into the PDE. I got $$ A'' - 2ik A' - k^2 A - \ddot{A} - 2i\omega \dot{A} + \omega^2 A = \lambda |A|^2 A \, . $$ Then, one could say that any change is slow, hence $A'' \to 0$ and $\ddot{A} \to 0$. Also, one could assume that, for a particular $t$, $A$ does not change for $x$, hence $A' \to 0$. What's left is $$ (\omega^2 - k^2) = \lambda |A|^2 + 2i\omega \frac{\dot{A}}{A} \, . $$ That looks like a dispersion relation, where the $\lambda$ term produces dispersion and the $\dot{A}$ acts as damping.

Do you have any ideas how to continue? I would prefer something general, but if not possible, you can assume $\lambda$ is small. Also, physical insights into what's small would also be appreciated.

Thanks a lot. SSF


1 Answer 1


Self-similar solutions

Notice that the the solution is invariant under the one-parameter transformation $$\phi(x,t)\mapsto \mu \phi(\mu x, \mu t), \mu\in\mathbb R^+$$ which we can interpret as an action of a Lie group, as well as translations in time, which means we can put $\mu=\frac{1}{T-t}$ to obtain self-similar solutions $\phi(x,t)=\frac{1}{T-t} f(\xi)$, where $\xi:=\frac{x}{T-t}$. $T$ is some characteristic "blowup" time.

After a slightly laborious calculation, the original equation transforms into an ODE, $$(1+\xi^2)f''(\xi)-4\xi f'(\xi)-2f(\xi)+\lambda [f(\xi)]^3=0.$$ It's formidable, but at least admits a constant solution $f(\xi)=\sqrt{2/\lambda}$.

Other solutions grow in $\xi$. For instance, here is the plot and the phase portrait of the solution with data $(f(0)=1,f'(0)=0,\lambda=1)$, crammed up to conserve some space!

enter image description here

Note that here the you need to interpret $\xi$ carefully. Moreover, the 2nd order ODE may be recast into a system of 1st order ODEs $$\frac{d f}{d\xi}=v, \:\frac{d v}{d\xi}=\frac{f(2-\lambda f^2)+4\xi v}{1+\xi^2}$$ The system is nonautonomous but we may cheat by analysing its behaviour at a fixed $\xi$. The critical points where the LHS's vanish are $(f,v)=(0,0)$ and $(\sqrt{2/\lambda},0)$, for all $\xi$. The eigenvalues turn out to be positive and negative for all $\xi$, meaning it's a saddle point, and complex but with $\Re{\lambda_{\pm}}>0$, meaning it's again unstable. Therefore all such solutions blow up.

Travelling solutions

We set $\eta=x-ct$ and try out solutions of the form $\phi(x,t)=g(\eta)$. With this the equation is recast as $$g''(\eta)=\tilde\lambda [g(\eta)]^3$$ where $\tilde\lambda:=\lambda/(c^2-1)$. Similarly as with trig/hyperbolic functions, this either blows up or oscillates, depending on the sign of $\tilde\lambda$. For negative $\tilde\lambda$, a solution is given in terms of the Jacobi elliptic function

$$g(\eta)=A\,cn( A \sqrt{-\tilde\lambda}\eta,1/\sqrt{2})$$

This solution is known as a compact breather. It's periodic, but probably highly unstable. The velocity and amplitude are probably independent parameters.

enter image description here

So we can try the ansatz $\phi(x,t)=A(x,t)e^{i(x-ct)}$ which leads to, as you have observed, $$(1-c^2)A_x+A_{tt}-A_{xx}+2i(A_x-c A_t)=\lambda |A|^2 A,$$ which is more difficult than the original equation unless further assumptions are made. If we insist on $A(x,t)$ being real, this implies $A_x=c A_t$. We want to avoid having an amplitude that is also of the form of a travelling wave, since that leads to the equation in the previous paragraph. We can check this by trying an ansatz $\phi(\eta,t)=A(\eta)e^{i(n \eta -mt)}$, $\eta=x-ct$, giving us

$$c^2 A''(\eta)-2 i c (-c m-n) A'(\eta)-(-c m-n)^2 A(\eta)-A(\eta)=\lambda |A(\eta)|^2 A(\eta)$$

where we can choose $cm+n=0$ to eliminate the $A'$ term, implying $A(\eta)$ is real, and giving us

$$c^2 A''(\eta)=\lambda A(\eta)^3+A(\eta)$$

This can be integrated if we mulitply by $ 2/c^2 \cdot A'$ giving us

$$A'^2=(\alpha_0+\alpha_1 A^2)(\beta_0+\beta_1 A^2)\Rightarrow \eta=\int^{A(\eta)}_0 \frac{\mathrm d \tilde A}{\sqrt{(\alpha_0+\alpha_1 \tilde A^2)(\beta_0+\beta_1 \tilde A^2)}}$$

which is essentially the same as before, the solution being given in terms of a Jacobi elliptic function as $$A(\eta)=\sqrt{\frac{\alpha_0}{\alpha_1}} sn(\sqrt{\alpha_1\beta_0}\eta, (\alpha_0\beta_1)/(\alpha_1\beta_0))$$

Linearisation; Stable and self-modulating solutions

Let's again try the ansatz $\phi(x,t)=A(x,t)e^{i(\omega_0 t - k_0 x)}$. I will change the original equation slightly, into $\phi_{tt}-c^2\phi_{xx}=\lambda |\phi|^2\phi$, where $c=\omega/k$ is the natural phase velocity of the free linear wave:

$$A_{tt}-c^2 A_{xx}+2 i (c^2 k_0 A_x+\omega_0 A_t)+(c^2k_0^2-\omega_0^2)A-\lambda |A|^2 A=0$$ introducing group velocity $c_g=c^2\frac{k_0}{\omega_0}=\frac{d\omega}{dk}\big|_{k=k_0}$ and a coupling parameter $\alpha=\lambda/(2\omega_0)$ $$\frac{1}{2\omega_0}(A_{tt}-c^2 A_{xx})+ i(c_g A_x+ A_t)+\frac{\omega_0}{2}\left((\frac{c_g}{c})^2-1\right)A-\alpha |A|^2 A=0$$ which we can further simplify by putting $\rho=\frac{\omega_0}{2}\left((\frac{c_g}{c})^2-1\right)$.

Now we switch to coordinates moving at group velocity, $\xi=x-c_g t,\tau=t$, so that we are essentially looking at how the envelope behaves:

$$i A_{\tau}-\beta A_{\xi\xi}+\rho A-\alpha |A|^2 A = 0$$

where $\beta=c_g^2-c^2=\frac{2 c^2}{\omega_0}\rho= \frac{1}{2}\frac{d^2\omega}{dk^2}\big|_{k=k_0}$ is a dispersive parameter. This is a nonlinear Schrodinger equation with a linear and cubic potential.

We can further suppose $A(\xi,\tau)= a(\xi,\tau)e^{i \theta(\xi,\tau)}$ which gives us two coupled equations

$$a_{\tau}-2\beta a_{\xi}\theta_{\xi}-\beta a \theta_{\xi\xi}=0$$ $$\rho a -\alpha a^3+\beta a \theta_{\xi}^2- a \theta_{\tau}-\beta a_{\xi\xi}=0$$

which we can slightly rearrange (we divide the second one with $a$ and differentiate wrt $\xi$)

$$(a^2)_{\tau}-2\beta(a^2\theta_{\xi})_{\xi}=0$$ $$(\theta_{\xi})_{\tau}-2\beta(\theta_{\xi})(\theta_{\xi})_{\xi}=-\alpha(a^2)_{\xi}-\beta (a_{\xi\xi}/a)_{\xi}$$

This strongly suggests putting $\Theta=-2\beta\theta_{\xi}$: $$(a^2)_{\tau}+(a^2\Theta)_{\xi}=0$$ $$\Theta_{\tau}+\Theta\Theta_{\xi}=-2\alpha\beta(a^2)_{\xi}+2\beta^2 (a_{\xi\xi}/a)_{\xi}$$.

The local frequency and wavenumber are described by $\omega=\omega_0+\phi_t$, and $k=k_0-\phi_{\xi}=k_0+\frac{1}{2\beta}\Theta$. If we consider small modulations we can safely discard the $(a_{\xi\xi}/a)_{\xi}$ term. Still, we see that frequency and amplitude modulations are not independent from each other. We can linearize the equations near harmonic solutions with $a=a_0$ and $\Theta=0$ so that $a=a_0+\tilde a$ and $\Theta=\tilde\Theta$ where $|\tilde a|<<a_0$, $|\tilde\Theta|<<1$. Plugging it in:

$$a_0\tilde\Theta_{\xi}-2\tilde a_{\tau}=0$$ $$\tilde\Theta_{\tau}+4\alpha\beta a_0 \tilde a_{\xi}=0$$

We try solutions of the system with $\tilde\Theta,\tilde a$ proportional to $\exp[i(\Omega\tau-\kappa\xi)]$ and solve the resulting linear system of equations for the amplitudes $\tilde \Theta_0,\tilde a_0$:

$$-\kappa a_0 \tilde \Theta_0 - 2 \Omega \tilde a_0 = 0 $$

$$\Omega \tilde \Theta_0 - 4 \alpha\beta a_0 \kappa \tilde a_0 = 0 $$

and the vanishing of the determinant gives a dispersion relation: $$\Omega^2=2\alpha\beta a_0^2\kappa^2$$

Finally, we see that we are dealing with either stable or unstable solutions, depending on the sign of $\alpha\beta$, which is the product of the dispersive term $\beta=c_g^2-c^2=\frac{1}{2}\frac{d^2\omega}{dk^2}\big|_{k=k_0}$ and the nonlinear cubic coupling term $\alpha=\lambda/(2\omega_0)$. If the product is positive, we are dealing with stable waves. This is essentially the case with the Jacobi elliptic waves we encoutered earlier, where we wanted $\tilde\lambda=\lambda/(c^2-1)$ to be negative. The notation is a bit off, as there $c$ was our $c_g$, and $c=1$.

On the other hand, for $\alpha\beta<0$ we have imaginary frequencies, and solutions $\exp{\pm \Omega \tau}$, which explodes as $t=\tau\rightarrow\infty$! Here the modulated wave interacts with the carrier wave and instability occurs.

Unfortunately I doubt I could say more about this wave equation! Perturbative solutions in $\lambda$ aren't that viable due to the cubic nonlinearity. The zeroth term would just reproduce an ordinary travelling wave, while the first term would again be complicated. I hope this was helpful!


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.