Problem
Following a precedent question, I realize that my thesis spacing problems are due to different packages in different condition. This time, it seems that the geometry package—or more precisely its configuration—changes spacing around equations. If we look at the rendered image, the spacing around the first equation (which has a smaller bounding box height) is higher than the other. On the right of the image is the output with the lua-visual-debug package. Note that this spacing problem is not specific to LuaLaTeX, and seems to disappear when the geometry setup line is removed. Any idea on what might cause the problem?
MWE
\documentclass[a4paper,oneside,12pt]{book}
\renewcommand{\baselinestretch}{1.2}
\usepackage[headheight=15pt]{geometry}
\geometry{hmargin=2.5cm, tmargin=3.5cm, bmargin=3cm} % removing this line fixes (in this instance) the spacing
\let\vec\mathbf
% \usepackage{lua-visual-debug}
\begin{document}
Une dernière modification de nos coordonnées consiste à se placer dans un référentiel rigidement lié aux atomes dans le laboratoire qu'on appelle référentiel mobile. On choisit le centre de gravité $G$ du système comme origine du repère, le vecteur $\vec{S}\lambda$ comme notre axe $z$, et le vecteur $\vec{s}\lambda$ situé dans le plan $xz$. On utilise alors les angles hypersphériques
\begin{equation}
\sigma_\lambda = (\omega_\lambda, \eta_\lambda, \phi_{S_\lambda}, \theta_{S_\lambda}, \varphi_\lambda)
\end{equation}
avec l'angle $\eta_\lambda$ définit comme
\begin{equation}
\eta_\lambda = \arccos \frac{\vec{s}\lambda \cdot \vec{S}\lambda}{s_\lambda S_\lambda}
\end{equation}
où l'angle $\varphi_\lambda$ est choisi tel que $\vec{s}_\lambda$ soit dans le plan $xz$.
\end{document}

geometry. It "looks" different because withoutgeometrythere is only a half-filled line which makes the spacing appear greater, but if you measure the distance between baselines it is exactly the same. – campa Feb 09 '21 at 07:56