When theta=0.5 then the exponent equals theta/(theta-1)=-1. The only nonlinearities in the model are of the form
xi * (gamma/xi)^(-1) = xi^2 / gamma
which is known in most modeling tools as quad_over_lin and the constraint
something >= xi^2/gamma
is equivalent to
something * gamma >= xi^2
and can be written using the rotated quadratic cone (assuming/enforcing gamma>0). So the model should be possible.
(I am assuming all coefficients like p_i are nonnnegative and the assummed gamma_i is nonnegative etc. as enforced by the remaining convexity requirements of the problem and by the definition of the RSOC cone).
From the statement "I tried to use mf.Domain.inRotatedQCone() this but it seems like it doesnt work" it is not clear what you did, what went wrong, or how to help.
Michal