[Navier$tokes11]

I've been using RANS turbulence models to try to replicate some sco2 heat exchanger performance. In general the models overpredict the heat transfer, which appears to be very common according to the literature. A few papers suggest this issue has to do with the sensitivity of the fluid properties, and specifically the variability of the turbulent prandtl number in different regions of the boundary layer. I'm wondering how one goes about determining a turbulent prandtl number? When experimental measurements aren't an option, is the only way to go about this to run a full fledged LES or DNS simulation and back out the value from there? Any advice would be appreciated!