Matthew Lee
Dr. Bennett
Article Summary 5
1. Bibliography
“[CFD] The Smagorinsky Turbulence Model (Part 1)” YouTube, uploaded by Fluid
Mechanics 101, 3 April 2021, https://youtu.be/V8ydRrdCzl0?feature=shared.
2. Reason you chose the article.
This video was a natural progression from the other videos I watched explaining some of
the theory behind RANS and LES modeling. I came across it while trying to understand different
turbulence models to eventually pick one to use for my project and this video was essential in
helping me reach that point in my understanding. Also, Jolie used this turbulence model in her
project, so that helped ground the model in practicality even more.
3. How did you find it?
I came across this Youtube channel and their series of videos on turbulence modeling
while looking for resources to better understand the models that I was using in OpenFOAM. The
videos on the Fluid Mechanics 101 channel in particular piqued my interest due to their
comprehensiveness and digestibility despite maintaining mathematical rigor throughout most of
the derivations and discussions.
4. Summary of the article. What new information did you learn?
From my last article summary, we know that an LES model will include an additional
sub-grid stress term τsgs to mimic the natural dissipation of eddies that are unable to break down
due to the finite mesh size. This τsgs is modeling using an eddy viscosity approach which involves
the sub-grid kinematic viscosity 𝜈sgs which needs to be calculated in order to close the system.
In the Smagorinsky model, a dimensional argument is made to find 𝜈sgs. The units of 𝜈sgs
are square meters per second, which is observed to be the same units as if we multiplied a
velocity in meters per second by a length in meters. As a result, we can find an expression for 𝜈sgs
by multiplying some characteristic velocity scale with some length scale for the sub-grid eddies
that compose the effects of 𝜈sgs.
For our length scale, we seek a length that encapsulates the effects of the sub-grid eddies.
First assume isotropic turbulence such that the eddies of the same size behave the same
regardless of their shape or orientation. This lets us characterize the sub-grid eddies with just
their size and allows us to neglect any additional geometric considerations that would complicate
our characterization. Since we want to capture the effects of all the sub-grid eddies, consider an
“average” or equivalent eddy that has the same turbulent kinetic energy as the average turbulent
kinetic energy of all sub-grid eddies. The diameter of such an eddy will be smaller than the mesh
size by the definition of a sub-grid eddy, and so we can assert that our characteristic length scale
can be calculated with
𝑙0 = 𝐶𝑠∆,
where ∆ is the mesh size and Cs is some constant between 0 and 1.
One reasonable choice for the characteristic velocity scale of a turbulent eddy is the
velocity difference across the eddy. This can be calculated as the product of the velocity gradient
across the eddy multiplied by its diameter or length scale. To incorporate the velocity gradients
in all directions in 3D space, we use the strain rate tensor defined by
1 ∂𝑈 ∂𝑈𝑗
𝑆𝑖𝑗 = 2
( ∂𝑥 𝑖 + ∂𝑥𝑖
).
𝑗
However, simply taking the magnitude of this tensor is insufficient as doing so would actually
underestimate the velocity gradients across our eddy by a factor of 2. We account for this by
multiplying our tensor magnitude by 2. Since our computational domain is discretized into a
mesh, it is fair to assume that the strain rate tensor evaluated for a cell is a good enough
approximation to characterize all sub-grid eddies within that cell. Therefore, our characteristic
velocity scale can be calculated with
𝑈0 = 𝑙0 · 2𝑆𝑖𝑗𝑆𝑖𝑗,
the product of the calculated velocity gradient and the length scale.
Recall that we are trying to model 𝜈sgs as the product of some characteristic velocity and
length scale. Using the equations derived above, this means that we model 𝜈sgs as
2
𝜈𝑠𝑔𝑠 = (𝐶𝑠∆) 2𝑆𝑖𝑗𝑆𝑖𝑗.
All that is left is to determine a value for the coefficient Cs. In the case of homogeneous isotropic
turbulence, Lilly (1966) analytically derived a value of Cs = 0.173. However, this value is
inaccurate in regions of high shear and near walls. To account for near-wall behavior, we can
either correct the length scale using a method like Van Driest damping or correct the velocity
scale using a method like WALE.
5. How does this new information relate to previous knowledge?
The video I summarized here is the first of two parts explaining the Smagorinsky LES
model. Since I had already learned about RANS models like the k-⍵ SST model, learning about
the Smagorinsky model completed my fundamental understanding of turbulence modeling using
both RANS and LES. It mostly built off concepts from the last source I summarized about LES
modeling in general, but it also used the strain rate tensor that I learned about from the Feynman
lectures in my fourth article summary.
6. Your thoughts about the new information.
This video and the next video were critical in the creation of my simulation model and
introduced me to the idea of van Driest damping, which greatly improved the accuracy of my
simulations. Altogether, the videos from this Youtube channel were invaluable in moving my
project forward and helping me get a grasp on the mathematics behind the models I am using.
