I was in for a surprise. Knowing that there are no dipole (or, for that matter, monopole) gravitational waves, I was absolutely convinced that an axisymmetric system -- say, a pair of black holes in a head-on collision -- cannot possibly generate gravitational waves. Simple symmetry arguments, I thought, clearly show that generation of quadrupole (or higher multipole) radiation would require breaking the axial symmetry.

I was wrong, and there is literature to prove it.

To see why, I figured I should construct a simple example: two weights, connected by a spring, oscillating lengthwise without rotation. Simple enough, is it not.

axis m m r(t) COM

Classical mechanics

The physics of this system in the context of classical mechanics is straightforward. Assuming that the two weights move around the $z$ axis and the spring is an ideal spring with relaxed length $L_0$ and spring constant $k$, the Lagrangian is

\begin{align}
L=\tfrac{1}{2}m\dot{z}_1^2+\tfrac{1}{2}m\dot{z}_2^2-\tfrac{1}{2}k(z_2-z_1-L_0)^2.\tag{1}
\end{align}

The corresponding Euler-Lagrange equations are, of course,

\begin{align}
\frac{\partial L}{\partial z_1}-\frac{d}{dt}\frac{\partial L}{\partial\dot{z}_1}=k(z_2-z_1-L_0)-m\ddot{z_1}&{}=0,\\
\frac{\partial L}{\partial z_2}-\frac{d}{dt}\frac{\partial L}{\partial\dot{z}_2}=k(z_1-z_2+L_0)-m\ddot{z_2}&{}=0.\tag{2}
\end{align}

A general solution is given in the form,

\begin{align}
z_{1,2}=\pm\tfrac{1}{2}L_0\pm\tfrac{1}{2}A\cos(\omega t+\phi),\tag{3}
\end{align}

i.e., symmetric oscillation with amplitude $A$, frequency $\omega=\sqrt{2k/m}$ and initial phase $\phi$ at $t=0$.

Gravitational radiation

To model gravitational radiation, we begin with Minkowski spacetime $\eta_{\mu\nu}$, slightly perturbed by $|h_{\mu\nu}|\ll 1$:

\begin{align}
g_{\mu\nu}=\eta_{\mu\nu}+h_{\mu\nu}.\tag{4}
\end{align}

The standard route then is to define the trace-reversed perturbation $\bar{h}_{\mu\nu}=h_{\mu\nu}-\tfrac{1}{2}\eta_{\mu\nu}h$ and impose the Lorenz gauge condition $\partial^\mu\bar{h}_{\mu\nu}=0$, which then leads to the linearized Einstein field equations in the form of a sourced wave equation,

\begin{align}
\Box\bar{h}_{\mu\nu}=-\frac{16\pi G}{c^4}T_{\mu\nu},\tag{5}
\end{align}

with the conserved stress-energy tensor, $\partial_\mu T^{\mu\nu}=0$, acting as source. This leads to the retarded Green's function solution

\begin{align}
\bar{h}_{\mu\nu}(t,\mathbf{x})=\frac{4G}{c^4}\int d^3 x' \frac{T_{\mu\nu}(t-|\mathbf{x}-\mathbf{x}'|/c,\mathbf{x}')}{|\mathbf{x}-\mathbf{x}'|}.\tag{6}
\end{align}

We explore the far field, $|\mathbf{x}-\mathbf{x}'|\approx|\mathbf{x}|=R$, leading to

\begin{align}
\bar{h}_{\mu\nu}(t,\mathbf{x})\approx\frac{4G}{c^4R}\int d^3x' T_{\mu\nu}(t-R/c,\mathbf{x}').\tag{7}
\end{align}

In the slow motion limit, $v\ll c$, $T^{00}\sim \rho c^2$, $T^{0i}\sim \rho cv^i$, $T^{ij}\sim \rho v^i v^j$, defining the quadrupole moment $I_{ij}(t)=\int d^3x \rho(t,\mathbf{x})x_ix_j$, we have

\begin{align}
\frac{d^2 I_{ij}}{dt^2}=2\int d^3x T^{ij}+{\cal O}(v^2/c^2),\tag{8}
\end{align}

allowing us to trade the integral for $\ddot{I}_{ij}$. In the far zone, this gets us the leading-order waveform

\begin{align}
h_{ij}^\text{TT}(t,\mathbf{x})=\frac{2G}{c^4R}\ddot{Q}_{ij}^\text{TT}(t-R/c),\tag{9}
\end{align}

where $Q_{ij}=I_{ij}-\tfrac{1}{3}\delta_{ij}\text{tr}~I$ is the trace-free quadrupole.

The corresponding energy-flux, averaged over several wavelengths, yields a well-known formula:

\begin{align}
P=\frac{G}{5c^5}\langle\delta^{ik}\delta^{jl}\dddot{Q}_{ij}\dddot{Q}_{kl}\rangle.\tag{10}
\end{align}

The case of the two-mass system

The standard Newtonian symmetric trace-free (SFT) mass quadrupole moment for the two-mass system is

\begin{align}
Q_{zz}=\tfrac{1}{3}mr^2(t),\qquad Q_{xx}=Q_{yy}=-\tfrac{1}{6}mr^2(t),\tag{11}
\end{align}

with $z_{1,2}=\pm\tfrac{1}{2}r$, thus $r=L_0+A\cos(\omega t+\phi)$.

Plugging this in, time-averaging $(\langle\sin(\omega t+\phi)^2\rangle=1/2, \langle\sin(\omega t+\phi)\rangle=0)$ and using $L_0\gg A$, we get

\begin{align}
P=\frac{G}{15c^5}\,m^2L_0^2A^2\omega^6.\tag{12}
\end{align}

Given the internal energy of the oscillator, $E=\tfrac{1}{4}m\omega^2 A^2$, and

\begin{align}
P=-\dot{E}=-\tfrac{1}{2}m\omega^2 A\dot{A},\tag{13}
\end{align}

we can solve for $A$:

\begin{align}
A(t)=A(0)e^{-t/\tau},\qquad \tau=\frac{15c^5}{2GmL_0^2\omega^4}.\tag{14}
\end{align}

That is an exceedingly small time constant. (For a system with two 1-kilogram masses 1 meter apart oscillating at 1 Hz, it takes approximately $4\times 10^{32}$ times the present age of the universe for the amplitude to decrease by one $e$-fold.) But it is not zero. The axisymmetric system radiates.

Further calculation also reveals that the radiation is exclusively of the "+" polarization and its magnitude is proportional to $\sin^2\theta$, where $\theta$ is the polar angle.

In light of this, I am no longer shocked or surprised to learn that both a test particle falling into a black hole [1] and two black holes colliding [2] head on radiate. But the point is, no black holes, no strong fields, no event horizons are needed. Gravitational radiation, however weak, is present even in this simple case of a nonrelativistic oscillator. Axisymmetry constrains the form of the quadrupole tensor but does not force it to be constant in time.


References

[1] Davis, M. et al., Gravitational Radiation from a Particle Falling Radially into a Schwarzschild Black Hole, PRL 27(21) 1971
[2] Anninos, P. et al., Collision of Two Black Holes, PRL 71(18) 1993