Wrong offset when using math mode and subscripts in Matplotlib and OSX

I have some plots where I would like to use subindices in some labels but whenever I use math mode all the labels appear shifted. Can I set some offset for all my labels? Is there anything I'm missing for using math mode?

This is the plot without math mode in the labels:

And this is what it looks like with math mode (notice the ticks):

For reference, here is my full code (I got the stacked code from ):

import numpy as NP
import matplotlib.patches as mpatches
import matplotlib.pyplot as plt
import matplotlib.cm as cm
data = '''0    0    0    0    0    0    0    0
0    0    0    0    0    0    0.015    0.015
0    0    0    0    0    0    0    0
0    0    0    0    0.04    0.04    0    0
0    0    0    0    0.03    0.01    0.19    0.14
0    0    0.772    0    0.07    0.01    0.12    0.11
0    0.879    0    0    0    0.07    0    0.085
0.056    0    0    0    0    0    0    0
'''.splitlines()
data = tuple(reversed([NP.array([float(j) for j in i.split('    ')]) for i in data]))
colors = cm.rainbow(NP.linspace(0, 1, 8))
axes = plt.figure().add_subplot(111)
axes.set_xticklabels([r'$m_%d$'%i for i in ([i+1 for i in range(8)])])
plt.stackplot(NP.arange(8)+1,
          data, 
          colors=colors)
plt.xlim(1,8)
plt.ylabel("Error")  
plt.legend([mpatches.Patch(color=i) for i in colors], 
           [r'$m_%d$'%i for i in ([i+1 for i in range(8)])])
plt.show()

Update: The problem resided on the backend used for interactive display

Following the hints provided in the comments I tried writing to a file and the labels appear properly. The problem seem to be on the MacOSX backend.

• Python 2.7.9 (default, Dec 11 2014, 02:36:08) [GCC 4.2.1 Compatible Apple LLVM 5.1 (clang-503.0.40)] on darwin
• matplotlib.version 1.4.3
• matplotlib.get_backend() MacOSX

There seems to be a bug on the MacOSX backend for Matplotlib. In order to solve it I had to switch the backend. I tried several from the FAQ (http://matplotlib.org/faq/usage_faq.html#what-is-a-backend) and I got best results using WXAgg. TkAgg was very sluggish and WX does NOT support math mode. If anyone is interested, the code to be added before importing pyplot is:

import matplotlib
matplotlib.use('WXAgg')

This are the results (all look slightly different):

WXAgg

TkAgg

WX