Introduction
Low Prandtl number liquid metals serve as primary coolant for MYRRHA1 and ALFRED2, two Gen-IV reactors under development1. The low Prandtl number induces discrepancies in the modeling of the turbulent heat transfer when directly addressed according to the Reynolds analogy with a constant turbulent Prandtl number. The thermal boundary layer is considerably larger than the velocity one, up to the point that, while we can clearly define a bulk velocity, the temperature profile may not exhibit any almost constant bulk temperature plateau. The issue was investigated by several authors who reviewed the existing correlations and proposed their own one2–4. The correlations make use of adimensional numbers such as Reynolds and Peclet. Their functional form is mostly empirical with coefficients determined by best fit. The Reynolds number and the Peclet numbers are global parameters, only useful and well-defined for simple geometries. Their use in CFD3 simulations of complex geometries is questionable. Among all correlations reviewed, Kays’ correlation is the only one making exclusive use of local parameters. Variants of this correlation have been used with significant success. The variants share the same structure but feature different values for one numerical constant. In our previously published four page brief report5, we showed that the correlations can be simply derived on a basic assumption with regards to the non-linear combination of stochastic effects and the variants then stem from different approximations of a mother formula. The objective of this former brief report was only to establish and keep trace of the probable relationship between the stochastic concept and the empirical correlation. It is in no way a bulletproof demonstration and there is no specific treatment of the viscosity. In particular, the definition of the asymptotic Prandtl number and its use are not completely consistent. Besides, from the purely numerical point of view, a defect of the mother formula, which is transferred to the variants, is that the turbulent Prandtl number becomes infinite at vanishing turbulence.
These considerations motivate to proceed further with the analysis which is the object of this current brief report. The key point resides in refining the concept that lies behind the formerly loosely determined asymptotic Prandtl number. What really needs to be done is to separate clearly what is from molecular origin and what is not. This is found to be more prolifically reformulated in terms of differentiating the properties of the fluid from the properties of the flow, being the flow either laminar or turbulent.
In order to proceed consistently with the flow versus fluid properties separation, it becomes necessary to apply the principle of square additivity not only to the thermal conductivity but also to the viscosity. This allows and brings us to define the ”flow Prandtl number”. Our driving hypothesis is that this number is a universal constant.
As previously, the principle of square additivity completely determines the turbulent Prandtl number. The new formula, while similar, is more articulated than the previous one presented in 5. Its first order approximation still has the functional shape of Kays’ correlation and coincides with it for low Prandtl number fluids. Our formula however gives additional information on Kays’ correlation range of validity. The new formula also correcly degenerates to the 0.85 traditional constant value commonly used for near unity Prandtl number fluids. Besides, as a good news for numerical implementation, the newly derived formula has the merit of not degenerating anymore at vanishing turbulence, which was not a feature of the previous formula.
Turbulent Prandtl number derivation
In the framework of thermal fluid dynamics of turbulent flows, the focus is concentrated on the determination and modeling of the effective viscosity and effective heat diffusion of the fluid. The viscosity and the heat diffusion both consist in the sum of two parts, one molecular and the other associated with turbulence.
With regard to the heat diffusion, both the molecular and the turbulent fluxes are oriented in the direction of the local temperature gradient and they are proportional to it. The intensity is also determined by the conductivity coefficients. The effective conductivity coefficient ke can be expressed as
denoting, respectively with k and kt the molecular conductivity and the increment due to turbulence.
Indicating with ρ the fluid density and Cp its specific heat, this expression can be rewritten in terms of thermal diffusivity:
in which αe = ke/(ρCp) stands for the effective diffusivity, α = k/(ρCp) is the molecular part and αt = kt/(ρCp) is the turbulent part.
Similarly, as far as the viscosity is concerned, the effective kinematic viscosity νe is the sum of the laminar contribution ν and the increment νt induced by turbulence:
The argument previously developed in 5 is the following. Conduction has a stochastic origin. The scale at which molecular conduction and the added conduction observed in turbulent flows operate are different and their mechanisms are unrelated. Molecular conduction is a molecular process and is a property of the fluid while the added conduction has a convective origin and is a property of the flow. The combination of their effects is better represented as a convolution rather than as a direct sum. Translated in formula, under this representation we expect to have:
where the diffusivity α0 should be an intrinsic property of the flow rather than the fluid.
The same argument is extended to the viscous process and it states that the effective viscosity comes from two independent processes whose intensity should be square additive:
By simple substitution, we have just defined two new quantities: the flow viscosity ν0
and the flow diffusivity α0
We can also redefine νt and αt in terms of ν0 and α0:
Therefore, the representations with (νt, αt) and with (ν0, α0) are completely equivalent and interchangeable.
We now introduce the known Prandlt number and turbulent Prandtl number . Besides, we can now also define the flow Prandtl number and look at how these numbers are related.
Noting that , by combining the former equations, we obtain:
This specific form is chosen to show that it cannot degenerate to zero or infinity. In effect, because ν0 goes to zero if νt does, then Prt tends to while for large turbulence, Prt tends to Pr0.
Expression 10 can be rewritten in terms of ν/ν0.
This formula is quite complicated and not easy to interpret at first glance, but considering ν/ν0 as a small parameter, a brutal simplification of 11 at first order in this parameter gives:
This formulation is practical only if ν0 is readily available. This would be the case if we had a transport equation for ν0 or a related variable similarly to what is done with the usual 2-equations turbulence models. Exploring the potential of this possibility is beyond the scope of the current discussion and we need to express Prt in terms of known parameters.
Thus, expressing Prt in terms of νt instead of ν0, formula 10 becomes:
this form being useful for interpretation at vanishing turbulent viscosity.
Dividing this last formula by Pr0, we have a relation between the three dimentionless groups , and in the form:
In view of a development in ν/νt, formula 13 becomes:
Here again, a brutal first order development in terms of ν/νt, valid only when both νt/ν and Prνt/Pr0ν are large, gives the following approximation for Prt:
meaning that changing from ν0 to νt brings only second order terms approximation. In particular, this last expression has the same form as the Kays correlation2.
Discussion
Our driving hypothesis is that the flow Prandtl number Pr0 is constant. Therefore, it must coincide with the value used for highly turbulent flow and near unit Pr. That is, Pr0 = 0.85.
For Pr ≃ 0.025, a typical value for heavy liquid metals, formula 16 becomes almost exactly the first Kays correlation:
For the second coefficient (here 0.7), Kays indicated two values, 0.7 and 2, discussing without reaching a conclusion in favor of one or the other value. However, a less brutal approximation could lead to a different coefficient. For example, in 6, 1.46 is obtained from direct analytical integration and best fit of heat transfer in a tube. Moreover, our newly derived formula is not much different than the one derived previously in 5 and a more precise approximation, while more complex to derive, would most probably also lead to an increased second coefficient about 1.45.
A particular case is when Pr = Pr0. Then Prt = Pr0 too. For medium and high Pr, we observe that Prt in formula (11) does not significantly departs from Pr0, except for strongly vanishing νt/ν.
With a little algebra, we found that Prt is a decreasing function of νt for Pr ≤ Pr0 and increasing function of νt for Pr ≥ Pr0, with values spanning the interval [Pr0; /Pr] and [ /Pr; Pr0]. The approximation in 16, while with the same monotonicity, fails to meet the correct bound for vanishing viscosity, for which it degenerates. Interestingly, we can see that the second coefficient depends critically on the Prandtl number, to the point that it vanishes for Pr = Pr0 and changes sign afterwards, like for the complete expression. This is a clear indication that the Kays correlation should be used as such only for low Prandtl number (say < 0.1) fluids.
The derived formula is ought to be used within a turbulence model. It would make sense only if the turbulence model correctly predicted the turbulent viscosity. This has to be true not only in the viscous boundary layer but also and principally in the bulk, in which the thermal boundary layer could still be in development. The problem is that the turbulence models mainly focus on the correct near-wall boundary layer turbulent viscosity profile, as it is the place where almost all the pressure drop is built, with the main aim to capture the correct wall shear stress. The turbulent viscosity profile in the bulk is normally of no practical importance, except for thermal flows of low Prandtl number fluids.
Looking at the profile of νt compared with direct numerical simulation data even in the simplest 2D channel flow, see figure 5 in 7, we can see that the turbulence models fail to give a correct profile for a wall Y+ value above 30. In particular, νt is underpredicted below Y+ up to 100–150 and over predicted afterward. The difference between the simply additive and the square additive approaches is mainly concentrated and felt around the values where both contributions are of similar intensity: ν/Pr ~ νt/Pr0 or equivalently νt/ν ~ Pr0/Pr. For a low Prandtl number fluid with Pr = 0.025 we have Pr0/Pr = 34, so all the region where νt/ν lies between say 10 and 100 is concerned, that is precisely for Y+ above 30 for the case analyzed in 7. The balance between the thermal effects of both Y+ regions is shifted towards an artificially increased diffusion. To counteract this effect, the turbulent Prandtl number can be increased artificially for a better fit. This could be an explanation for the use of an augmented second coefficient in the Kays correlation as indicated previously.
The main effect of applying the square additivity to the effective viscosity is that it removes the degeneration of the Prt formula for vanishing viscosity that was still present in 5. It is not clear nor sure that it is of practical importance anywhere else. It seems that, within the level of approximation given by the 2-equations turbulence models, νt and ν0 can be used indifferently in the formula. In other words, the square additivity could be used solely for the energy equation.
Conclusions
We consider that the effective viscosity takes origin from two independent stochastic processes whose intensity is square additive. The same consideration is extended to the effective conductivity. We have defined a flow Prandtl number which is expected to be a universal property of the flow and to be in fact a constant under the square additive approach. The turbulent Prandtl number is then determined by a formula reproducing the first variant of Kays’ correlation when approximated at the first order. The derivation sheds light on the Kays correlation and indicates that the second coefficient depends critically on the Prandtl number to the point that it vanishes when Pr = 0.85. Under the condition that the classical 2-equations turbulence models become able to capture correctly the turbulent viscosity profile, we expect that the turbulent Prandtl number formula can give improved thermal results independently of the Prandtl number and particularly for the low Prandtl liquid Lead and Lead alloys.
The Reynolds analogy could have a much wider domain of validity by combining it with the SquAd (Square Additive) derivation.
