= Solution
Take $0<v<1$ and $\Gamma=(1-v^2)^{-1/2}$. The <Sine-Gordon kink-antikink scattering solution> tends to the same vacuum at both spatial ends and hence has zero net <topological charge>. At large $|t|$, the two transition centers, found by setting the magnitude of the inverse-tangent argument to one, obey
$$
\Gamma|x|=\Gamma v|t|-\log v+o(1).
$$
They are a <kink> and an <antikink> moving with asymptotic speeds $\pm v$, with no outgoing radiation. Their left-right orientations interchange at the collision. To define the <soliton time delay>, label a transmitted trajectory by its preserved <rapidity> rather than its left-right position. The right-moving asymptotes are
$$
x_{\rm in}=vt+\frac{\log v}{\Gamma},\qquad x_{\rm out}=vt-\frac{\log v}{\Gamma}.
$$
After restoring physical lengths, the forward shift is $\Delta X=-2\log v/(m\Gamma)>0$. The arrival-time difference is therefore
$$
\boxed{\Delta T=-\Delta X/v=\frac{2\log v}{m\Gamma v}=\frac{2\log\tanh(\theta/2)}{m\sinh(\theta/2)}<0,\qquad v=\tanh(\theta/2).}
$$
This is a time advance relative to the extrapolated incoming free motion.
\b[The semiclassical scattering phase.] For an outgoing energy wave packet with full <S-matrix> phase $S=e^{i\delta(E)}$, <stationary phase> in $e^{-iET+i\delta(E)}$ shifts its arrival time by $d\delta/dE$. The <full S-matrix phase from a classical soliton delay> uses the paper's $\delta$, twice the phase in a convention $S=e^{2i\delta_{\rm pw}}$. In the center-of-momentum frame, $E=2\mathcal M_K\cosh(\theta/2)$, so $dE/d\theta=\mathcal M_K\sinh(\theta/2)$. It follows that
$$
\boxed{\frac{d\delta_{\rm sc}}{d\theta}=\frac{2\mathcal M_K}{m}\log\tanh(\theta/2).}
$$
The literal PDF normalization gives $\mathcal M_K=8m/\beta$ and consequently
$$
\boxed{\delta_{\rm literal}(\theta)=\delta_{\rm literal}(0)+\frac{16}{\beta}\int_0^\theta\log\tanh(u/2)du.}
$$
The integrable logarithmic singularity at zero causes no divergence of this phase difference. With the intended standard angular-field prefactor $m^2/\beta^2$, the result is instead
$$
\boxed{\delta_{\rm standard}(\theta)=\delta_{\rm standard}(0)+\frac{16}{\beta^2}\int_0^\theta\log\tanh(u/2)du.}
$$
The <soliton time delay> determines only phase differences, not an energy-independent constant or a choice of $2\pi$ branch. Keeping this distinction avoids an arbitrary high-energy subtraction.
\b[Bound-state poles.] Set $a_j=\pi j/N$. Each factor of the exact amplitude can be written $\cosh[(\theta-ia_j)/2]/\cosh[(\theta+ia_j)/2]$. Its denominator vanishes in the <physical rapidity strip> at
$$
\theta=iu_n,\qquad u_n=\pi(1-n/N),\qquad n=1,\ldots,N-1.
$$
The corresponding numerator is nonzero, and no other numerator cancels the pole. In the direct <kink>–<antikink> fusion interpretation, analytically continue the constituent <rapidities> to $\pm iu_n/2$. Their momenta sum to $(2\mathcal M_K\cos(u_n/2),0)$, so the <relativistic bound-state mass from a rapidity pole> gives
$$
\boxed{\mathcal M_n=2\mathcal M_K\sin\frac{n\pi}{2N},\quad n=1,\ldots,N-1.}
$$
These neutral particles form the <Sine-Gordon breather spectrum at reflectionless couplings>: there are $N-1$ breathers, none for $N=1$, and the putative $n=N$ state is at the unbound $2\mathcal M_K$ threshold. The amplitude fixes these ratios to the physical <kink> <mass>; its <rapidity> dependence alone cannot fix the overall <mass> scale or a mass-renormalization prescription relating that <mass> to $m$.
It is important to distinguish direct and crossed interpretations rather than count every occurrence of a pole twice. <Crossing symmetry> sends $iu_n$ to $i(\pi-u_n)=iu_{N-n}$. At the original angle the momentum-difference invariant is $4\mathcal M_K^2\sin^2(u_n/2)=\mathcal M_{N-n}^2$, so the crossed interpretation exchanges the complementary member of the same tower. In particular the transmission residue alternates sign: direct evaluation of the remaining factors gives $\operatorname{Res}_{iu_n}S_T=i(-1)^n R_n$, $R_n>0$. Thus one must retain charge-channel/crossed-channel information, not reject every negative-imaginary transmission residue or declare an additional particle for a crossed pole. Direct and crossed locations coincide in the reflectionless pole set. The familiar $N=2$ case has one breather of <mass> $\sqrt2\mathcal M_K$ despite the negative-imaginary transmission residue; this diagonal reflectionless example is also displayed in https://arxiv.org/pdf/1310.4779[Castro-Alvaredo, Chen, Doyon and Hoogeveen, section 4.2].
\b[Matching the exact phase.] For the <unwrapped reflectionless sine-Gordon transmission phase>, at real $\theta\geq0$ choose the continuous unwrapped branch with $\delta_N(0)=\pi N$. Then
$$
\boxed{\delta_N(\theta)=\pi N-2\sum_{j=1}^{N-1}\arctan\left[\tanh(\theta/2)\tan\frac{\pi j}{2N}\right].}
$$
Every factor has unit modulus, consistent with purely transmitting elastic scattering. Differentiating is simpler than integrating its complex logarithm:
$$
\delta_N'(\theta)=-\sum_{j=1}^{N-1}\frac{\sin(\pi j/N)}{\cosh\theta+\cos(\pi j/N)}.
$$
For fixed $\theta>0$ the sum becomes a <Riemann sum>, and the elementary integral yields
$$
\begin{aligned}
\delta_N'(\theta)&\sim-\frac N\pi\int_0^\pi\frac{\sin a}{\cosh\theta+\cos a}da\\
&=-\frac N\pi\log\frac{\cosh\theta+1}{\cosh\theta-1}
=\frac{2N}{\pi}\log\tanh(\theta/2)
=\frac{16}{\gamma}\log\tanh(\theta/2).
\end{aligned}
$$
Since $\beta^2=8\pi/(N+1)$ and $\gamma=8\pi/N$, replacing $\gamma$ by $\beta^2$ changes this only at subleading order. The leading phase difference agrees with the standard semiclassical result. The derivative approximation is not uniform as $\theta\to0$ at fixed $N$; integrating its logarithm gives a finite leading phase difference nevertheless. A compatible constant is the given branch $\pi N$, whose choice is supplied by the exact amplitude rather than the classical trajectory.
As a useful endpoint check, $\int_0^\infty\log\tanh(u/2)du=-\pi^2/4$, obtained by expanding the two logarithms $\log(1-e^{-u})-\log(1+e^{-u})$. Thus the leading phase with coefficient $2N/\pi$ tends to $\pi N/2$. The exact unwrapped endpoints are
$$
\boxed{\delta_N(0)=\pi N,\qquad\delta_N(+\infty)=\frac{\pi(N+1)}2,}
$$
where the extra $\pi/2$ is subleading. Neither should be replaced by zero by an unannounced branch convention. Finally, the literal $m^2/\beta$ action gives a phase derivative of order $\sqrt N$, whereas the printed exact amplitude gives order $N$. \b[The requested exact/semiclassical matching requires correcting the overall action normalization to $m^2/\beta^2$.] This is an actual inconsistency between pages 2 and 4, not a change to the sine-Gordon equation or the classical scattering field.
Back to article page
