Okay they are expanding the function
This function is diff at 0
if you expand this using the forumla above you will get
This is exactly the same series you get if you compose

with the taylor series I have above
The bad thing about this new series is that it does not alternate
So now we have to bound it using taylors remainder theorem (It is alot more work)
Since we are bounding the error on a degree 3 we need to find the maximum of the absolute value of the forth derivative.
we are interested in the value of
Since the fourth derivative is decreasing and negative on the interval

its max must occur at the right end point of at
and
So taylors bound on the remainder is
This is identical to the error given by AST estimate in my first post.
Rember is the series is alternating using the AST estimation is alot easier.
I hope this clears up the whole situation