Example

import matplotlib.pyplot as plt

kw = dict(
    fontsize=40,
    bbox=dict(boxstyle='round,pad=0', fc='lightsteelblue', ec='none')
)

plt.axhline(y=0.7)
plt.annotate('A', xy=(0.3, 0.7), **kw)
plt.annotate(r'$A$', xy=(0.4, 0.7), **kw)
plt.annotate(r'$A_{\tiny 1}$', xy=(0.5, 0.7), **kw)
plt.annotate(r'$A_{\tiny 1j}$', xy=(0.65, 0.7), **kw)

plt.axhline(y=0.4)
plt.annotate('A', xy=(0.3, 0.4), va='bottom', **kw)
plt.annotate(r'$A$', xy=(0.4, 0.4), va='bottom', **kw)
plt.annotate(r'$A_{1}$', xy=(0.5, 0.4), va='bottom', **kw)
plt.annotate(r'$A_{1j}$', xy=(0.65, 0.4), va='bottom', **kw)

Observations

• Having subscripts in LaTeX shifts the text upwards, but doesn't change the size of the bbox.
• Interestingly though, the box gets higher, if I use 1j instead of 1 as a subscript.
• Verticalalignment (bottom or baseline (the default)) results in the same positioning of the box, no matter what text is used.

@timhoffm's example now gives

so it seems like the baseline case has been fixed (possibly by #16476). The "bottom" alignment case remains, but I suspect that can be fixed "relatively" easily because the dvi parser knows the descent of each glyph, so should be able to return the text's bbox. (Although it seems like "bottom" alignment has the big space below even in the non-usetex, non-mathtext case?)

Except for $A_{\tiny 1}$ all examples still appear to be shifted by 1px upwards.

My (non-verified) guess is that the 1px shift is related to the "wiggly baseline" mentioned in #5414, which basically arises because mathtext/usetex texts (which are basically prerendered as bitmaps first) use their top-right corner as reference point, but there's some rounding associated with it if the text object gets antialiased past the top limit, which (handwavingly) leads to the baseline wiggling; instead, these bitmaps should use the baseline as reference y to avoid the problem.

This still seems to be a problem with more complex situations:

import matplotlib.pyplot as plt

kw = dict(
    usetex=True,
    fontsize=40,
    bbox=dict(boxstyle='round,pad=0', fc='lightsteelblue', ec='none')
)

plt.axhline(y=0.7)
plt.annotate('A \n B', xy=(0.4, 0.7), **kw)
plt.annotate(r'$A$' ' \n ' r'$B$', xy=(0.5, 0.7), **kw)
plt.annotate(r'$A_{\tiny 1}$' ' \n ' r'$B_{\tiny 1}$', xy=(0.6, 0.7), **kw)
plt.annotate(r'$A_{\tiny 1j}$' ' \n ' r'$A_{\tiny 1j}$', xy=(0.8, 0.7), **kw)

plt.axhline(y=0.3)
plt.annotate('A \n B', xy=(0.4, 0.3), va='bottom', **kw)
plt.annotate(r'$A$' ' \n ' r'$B$', xy=(0.5, 0.3), va='bottom', **kw)
plt.annotate(r'$A_{1}$' ' \n ' r'$B_{\tiny 1}$', xy=(0.6, 0.3), va='bottom', **kw)
plt.annotate(r'$A_{1j}$' ' \n ' r'$A_{\tiny 1j}$', xy=(0.8, 0.3), va='bottom', **kw)

plt.annotate(r'$^6\mathrm{He}$ $0^+$' ' \n ' r'$n0p_{3/2}$', xy=(0.05, 0.3), va='bottom', **kw)

plt.show()

gives:

This looks to be an issue with superscripts -- I will open a new issue.

Your example is helpful. Maybe add to #7075 which I think is the same thing (if more convoluted)

This issue has been marked "inactive" because it has been 365 days since the last comment. If this issue is still present in recent Matplotlib releases, or the feature request is still wanted, please leave a comment and this label will be removed. If there are no updates in another 30 days, this issue will be automatically closed, but you are free to re-open or create a new issue if needed. We value issue reports, and this procedure is meant to help us resurface and prioritize issues that have not been addressed yet, not make them disappear. Thanks for your help!

Haven't checked but almost likely still valid as no work has been done on these aspects, and will likely be stalled until the image comparison framework is fixed as any change to the wiggly baselines will involve regen'ing many, many reference images (#21653 (comment))?