%i arises when simplifying a square root in real domain

Dear all,

I am having trouble with maxima when using square roots of the form
sqrt(1-a**2), as in some situations they get simplified into 
%i*sqrt(a**2-1).

E.g., even with assume(a<1), if I try to get back to sqrt(1-a**2) from 
this modified form:

(%i20) ratsimp(1/sqrt(a**2/(1-a**2)+1));
                                          2
(%o20)                         - %i sqrt(a  - 1)

It is then very awkward, as rootscontract e.g. would make the mistake of 
writing %i=sqrt(-1), and get the wrong sign.

I could not find a boolean to tell maxima to avoid factoring out sqrt(-1), 
does it exist?

Thanks.

-- 
Jocelyn ?TIENNE                           Research Scientist of the CNRS      
Laboratoire Interdisciplinaire de Physique        Universit? de Grenoble  
Jocelyn.Etienne at ujf-grenoble.fr   www-liphy.ujf-grenoble.fr/link/etienne
On leave at U. Cambridge until Dec 2012,  www.pdn.cam.ac.uk/staff/sanson