Intermittent bug: Equation shifts to the left along with number [closed]

Question

I have an issue with an equation in my latex intermittently shifting to the left. It is really sensitive to the layout of the page in general -- which means that I can't get a truly minimal working example, so I'll just show snapshots. The code is included below. Here, you see the bad alignment.

Eq. (A10) is flush left, but the equation number is also moved to the left. It's just a plain old equation environment. I can change its content to anything, and the alignment is still bad, so it's not that.

If I change the rest of the page in any way (e.g., remove a couple lines from well before or well after), the alignment issue resolves. Here, I've just added a \newline to a paragraph far before the equation (not changing this paragraph in any way):

Obviously, I don't want that newline. Any ideas on where this might be coming from? Or hacks to try to get it working without an extra paragraph break?

I'm using revtex4-1 in reprint mode, with amsmath and about a hundred other things that I can't change without changing the layout of the article so much that this intermittent problem goes away. That is, I can't narrow it down by removing packages. Sorry this is so vague and unhelpful, but it's intermittent, so I can't figure out how to simplify it. One of you geniuses must have some idea!

Code

The SWSHs form a basis for spin-weighted functions on the\footnote{The
  choices of signs in these definitions are---to a great
  extent---arbitrary conventions.  However, care must be taken to
  ensure that the resulting $\mathfrak{D}$ matrices form a
  representation of the rotation group rather than an
  anti-representation, and to ensure that the handedness of space is
  preserved.  Our purpose in choosing these particular signs is to
  reproduce the standard SWSHs as special cases.  In particular, note
  that the presence of $Q_{3}$ in the definition of $Q_{a}$ is what
  picks out the $z$ axis as the point of reference on the sphere, so
  that angles are measured with respect to it, rather than the $x$ or
  $y$ axes.}  sphere~\cite{NewmanPenrose:1965, JanisNewman:1965,
  NewmanPenrose:1966}.  Goldberg \textit{et
  al.}~\cite{GoldbergEtAl:1967} showed that the SWSHs can be expressed
as special cases of Wigner's $\mathfrak{D}$ matrices, so that by
constructing $\mathfrak{D}^{(\ell)}_{m', m}$, we will obtain
${}_{s}Y_{\ell,m}$.  Defining the parts of the quaternion $\mathbf{Q}$
as
\begin{equation}
  x~,
\end{equation}
we can express quaternion multiplication as
\begin{gather}
  y\\z~.
\end{gather}
Then, following the standard derivation~\cite{*[{Chapter~15 of }] [{}]
  Wigner:1959}, we obtain