# Long equations break with split or multlined

I am using the IEEEtran class (two columns) for a paper, and have several long expressions that I cannot figure out the best way to format them. I am currently using \split and aligning everything manually as I see best for clarity, but this takes up a lot of space. Perhaps with the use of \multlined and \shoveright I can get better results. Below is the MWE:

\usepackage{mathtools,amssymb}

\begin{document}
\begin{align}
\begin{split}
\mathcal{UB}\left(\underset{\boldsymbol{Z}}{\mathrm{E}}\log C_{pq}\right)
&=-\frac{\log p}{p}-\frac{\log q}{q}\\
&\phantom{=}-\mathcal{E}_B\log m_{Z}-\log m_{X}\\
&\phantom{=}+ \mathcal{E}_B\log\left(m_{X}M_{X}^{q-1}+m_{Z}M_{Z}^{p-1}\right)\\
&\phantom{=}+\Pr{(B)}\log\Biggl( m_{X}M_{X}^{q-1}\\
&\phantom{=+\Pr{(B)}\log\Biggl(}
+\inf_{Z\in B}m_{Z}\left(\sup_{Z\in B}M_Z\right)^{p-1}\Biggr)\\
&\phantom{=}-\Pr{(B)}\log\inf_{Z\in B}m_{Z},
\end{split}
\end{align}
\end{document}


If using \multlined then the MWE would be:

\begin{align}
\begin{multlined}
\mathcal{UB}\left(\underset{\boldsymbol{Z}}{\mathrm{E}}\log C_{pq}\right)
=-\frac{\log p}{p}-\frac{\log q}{q}\\
-\mathcal{E}_B\log m_{Z}-\log m_{X}\\
+ \mathcal{E}_B\log\left(m_{X}M_{X}^{q-1}+m_{Z}M_{Z}^{p-1}\right)\\
+\Pr{(B)}\log\Biggl( m_{X}M_{X}^{q-1}\\
+\inf_{Z\in B}m_{Z}\left(\sup_{Z\in B}M_Z\right)^{p-1}\Biggr)\\
-\Pr{(B)}\log\inf_{Z\in B}m_{Z},
\end{multlined}
\end{align}


In my full document, as the IEEEtran is two column, the equations often overflow a bit into the other column. What would be the best way to handle these scenarios?

## First proposal

The math fonts clearly clash with the text font.

\documentclass{IEEEtran}
\usepackage{amsmath,amssymb}

\usepackage{lipsum}

\DeclareMathOperator*{\E}{E}

\begin{document}

\lipsum[1][1-4]
\begin{split} &\mathcal{UB}\bigl(\,\E_{\boldsymbol{Z}}\log C_{pq}\bigr) \\ &\quad=-\frac{\log p}{p}-\frac{\log q}{q}-\mathcal{E}_B\log m_{Z}-\log m_{X}\\ &\qquad+ \mathcal{E}_B\log(m_{X}M_{X}^{q-1}+m_{Z}M_{Z}^{p-1})\\[1ex] &\qquad\begin{aligned}[t] &+\Pr(B)\log\bigl( m_{X}M_{X}^{q-1}\\ &\qquad+\inf_{Z\in B}m_{Z}\bigl(\sup_{Z\in B}M_Z\bigr)^{p-1}\bigr) \end{aligned} \\[1ex] &\qquad-\Pr(B)\log\inf_{Z\in B}m_{Z}, \end{split}
\lipsum[2][1-4]

\end{document}


## Second proposal

Matching math and text fonts.

\documentclass{IEEEtran}
\usepackage{amsmath}
\usepackage{newtx}
\usepackage{bm}

\usepackage{lipsum}

\DeclareMathOperator*{\E}{E}

\begin{document}

\lipsum[1][1-4]
$$\begin{split} &\mathcal{UB}\bigl(\,\E_{\bm{Z}}\log C_{pq}\bigr) \\ &\quad=-\frac{\log p}{p}-\frac{\log q}{q}-\mathcal{E}_B\log m_{Z}-\log m_{X}\\ &\qquad+ \mathcal{E}_B\log(m_{X}M_{X}^{q-1}+m_{Z}M_{Z}^{p-1})\\[1ex] &\qquad+\Pr(B)\log\bigl( m_{X}M_{X}^{q-1} +\inf_{Z\in B}m_{Z}\bigl(\sup_{Z\in B}M_Z\bigr)^{p-1}\bigr) \\[1ex] &\qquad-\Pr(B)\log\inf_{Z\in B}m_{Z}, \end{split}$$
\lipsum[2][1-4]

\end{document}


## Comments

I don't think there's a gain in clarity with big fences. Actually, I believe the contrary.

The command \Pr doesn't take an argument. Look closely at the difference between \Pr(B) and \Pr{(B)} and you'll see that the latter is wrong.