Question: fsolve for real variables

I try to use fsolve to solve a set of equations for real parameters alphaij, thetha (they are the real angle and phases in the parametrization of a unitary 2x2 matrix)

assume(alpha11::real,alpha12::real,alpha22::real,theta::real);

myresult:=timelimit(10*time_max,fsolve({mat_M_V[2,1]=0,mat_M_V[1,2]=0,Im(mat_M_V[1,1])=0,Im(mat_M_V[2,2])=0},{alpha11=0,alpha22=0,alpha12=0,theta=0},'fulldigits'));

unfortunately fsolve ignores my assumptions and returns complex solutions

(alpha11 = -.20284e-19+.25703e-11*I, alpha12 = .10523e-19+.13389e-11*I, alpha22 = .20284e-19-.25703e-11*I, theta = 283.69+.37386e-8*I)

Obviously it thinks that the variables must be complex, just because the equations include complex numbers...

I tried to supply real ranges instead of the startvalues but in this case fsolve fails to find any solution. 

Also fsolve fails to distinguish between real and complex when it comes to the counting of variables and equations ...

I am using Maple 16. I suffer since several versions of Maple from the poor implementation of numerical root finding. The command has just too few options to influence its behaviour. Is there some alternative to fsolve within Maple or will this ever change (i.e. is there ongoing work on this command)? 

Any ideas appreciated!