I am doing calculations on small transition metal-oxide clusters with very small bandgaps using DMol3. The geometry optimization is possible, but only if I use thermal smearing. However, if I would like to calculate thermodynamic properties and for that, I start frequency calculations, they stop with SCF convergence error. After inspecting the output it seems that frequency calculation is 1) using numerical derivative calculations, and 2) all the calculations are without thermal smearing, therefore SCF is not converging even after small displacement.
How can I fix this? Is there a way to enforce thermal smearing during the frequency calculations (and if so, does it make sense)?