Standard $k\!-\!\varepsilon$ Model
Consider flows that satisfy the Boussinesq approximation (i.e., the fluid is approximately incompressible, and any influence of non-uniform temperature on the mean velocity field acts only through buoyancy). In this case, the exact transport equation for turbulent kinetic energy (the $k\!-\!\varepsilon$ equation) is $$ \frac{Dk}{Dt} =\frac{\partial}{\partial x_j}\!\left[ \nu\,\frac{\partial k}{\partial x_j} -\overline{\,u_j\!\left(\frac{u_i u_i}{2}+\frac{p}{\rho}\right)}\right] -\overline{u_i u_j}\,\frac{\partial U_i}{\partial x_j} -\beta\,g_i\,\overline{u_i\vartheta} -\varepsilon $$ where $D/Dt=\partial/\partial t+U_j\,\partial/\partial x_j$. $\overline{u_i u_j}$ is given by \(\boxed{-\overline{u_i u_j} =2\,C_\mu\,\frac{k^{2}}{\varepsilon}\,S_{ij} -\frac{2}{3}\,k\,\delta_{ij}}\), while $k$ and $\varepsilon$ are unknowns to be solved for. Only the turbulent diffusion term and the buoyancy production term on the right-hand side need to be modeled approximately
Referring to the laws of molecular diffusion and molecular heat conduction, the turbulent diffusivity of $k$ and the turbulent heat flux are approximated by the following gradient models $$ -\overline{\,u_j\!\left(\frac{u_i u_i}{2}+\frac{p}{\rho}\right)} =\frac{\nu_t}{\sigma_k}\,\frac{\partial k}{\partial x_j} \quad,\quad -\overline{u_i\vartheta} =\frac{\nu_t}{\sigma_t}\,\frac{\partial \Theta}{\partial x_i} $$ where $\Theta$ is the mean relative temperature; $\sigma_k$ and $\sigma_t$ are empirical constants ($\sigma_t$ is named the turbulent Prandtl number)

Substituting \(\boxed{-\overline{u_i u_j} =2\,C_\mu\,\frac{k^{2}}{\varepsilon}\,S_{ij} -\frac{2}{3}\,k\,\delta_{ij}}\), \(\boxed{-\overline{\,u_j\!\left(\frac{u_i u_i}{2}+\frac{p}{\rho}\right)} =\frac{\nu_t}{\sigma_k}\,\frac{\partial k}{\partial x_j}}\) and \(\boxed{-\overline{u_i\vartheta} =\frac{\nu_t}{\sigma_t}\,\frac{\partial \Theta}{\partial x_i}}\) into \(\boxed{\frac{Dk}{Dt} =\frac{\partial}{\partial x_j}\!\left[ \nu\,\frac{\partial k}{\partial x_j} -\overline{\,u_j\!\left(\frac{u_i u_i}{2}+\frac{p}{\rho}\right)}\right] -\overline{u_i u_j}\,\frac{\partial U_i}{\partial x_j} -\beta\,g_i\,\overline{u_i\vartheta} -\varepsilon}\), and neglecting molecular diffusion under high-Reynolds-number conditions, yields the $k\!-\!\varepsilon$ turbulence model \[ \frac{Dk}{Dt} =\frac{\partial}{\partial x_j}\!\left(\frac{\nu_t}{\sigma_k}\,\frac{\partial k}{\partial x_j}\right) +2C_\mu\,\frac{k^2}{\varepsilon}\,S_{ij}S_{ij} +\beta\,g_i\,\frac{\nu_t}{\sigma_t}\,\frac{\partial \Theta}{\partial x_i} -\varepsilon \] For high-Reynolds-number turbulence, the empirical constants are determined experimentally as $$ C_\mu=0.09,\qquad \sigma_k=1.0,\qquad \sigma_t=0.9 $$ Taking $C_\mu=0.09$ is a consequence of the local equilibrium characteristics of turbulent boundary layers.


1Song Fu & Liang Wang (2023). Theory of Turbulence Modelling. ISBN 978-7-03-074639-9.