The question about gmx h2order

GROMACS version: 2022.6
GROMACS modification: No

Dear all users,

I am doing an EOF flow case in a silica nanochannel. (y is height direction and z is length direction)

The water orientation and its angle to y axis is extracted by gmx h2order with this commend:
gmx h2order -f md.trr -s md.tpr -n index.ndx -o h2order.xvg -b 0 -e 10000 -d Y -sl 600
When I check the output xvg file, it is confusing that the output cosine value is different from the one I directly calculate from mu_x, mu_y, mu_z with the simple eqaution:
cosθ = (mu_y/sqrt(mu_x^2+mu_y^2+mu_z^2))

I try two different water model: SPC/E and TIP4P/2005.
It seems that the dependency between output cosine value and the calculated value is even larger for TIP4P/2005. (blue is from my calculation, orange is the angle calculated from the cosine value directly from h2order.)
SPC/E


TIP4P/2005

I am really grateful if anyone know the possible reason for it.
Thank you for your time and your help.